A method for testing the damage degree of active materials in lithium-ion batteries based on deep learning

Abstract

The present invention relates to a method for testing the degree of damage of active materials of lithium-ion batteries based on deep learning. The method comprises: preparing cross-section samples of lithium-ion battery pole pieces; collecting cross-section images with an optical microscope as original data sets; distinguishing and annotating original data set images as training data sets for deep learning models; building and training deep learning models; building an in-situ optical microscope observation system for active materials of lithium-ion batteries, and conducting in-situ observation experiments; inputting in-situ observation image results into the deep learning model to complete intelligent identification of particles of active materials of lithium-ion batteries to be observed; and determining the grayscale value of the image by setting a threshold value, thereby realizing quantitative statistical analysis of the degree of damage of lithium-ion battery materials. The present invention can perform automated and intelligent quantitative statistical analysis on the degree of damage of active materials of the same pole piece at different charge and discharge stages, and can be widely used in the study of the evolution mechanism of the degree of damage of lithium-ion battery materials.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion battery material characterization, and relates to a method for testing the degree of damage of lithium-ion battery active materials based on deep learning. Background Art

[0002] In the past few decades, lithium-ion batteries have become the technology of choice for a range of portable energy storage device solutions due to their high energy density and capacity. In order to meet the higher requirements of cycle life, power and energy density in emerging application scenarios such as green power grids and new energy vehicles, advanced lithium-ion battery technology still needs to continue to develop. The active materials of lithium-ion batteries are an important factor in determining the performance of lithium-ion batteries. During the charging and discharging process, lithium ions are embedded and released in the active materials, accompanied by the volume expansion and contraction of the active materials. Diffusion-induced stress is generated during the diffusion of lithium ions in the active materials. With long-term cyclic charging and discharging, lithium-ion battery materials will develop cracks of varying degrees, which will affect the performance of lithium-ion batteries. Therefore, exploring the crack evolution mechanism of lithium-ion battery materials has important guiding significance for optimizing the design of lithium-ion battery materials.

[0003] Usually, researchers use SEM to observe the cracks in the cross section of particles to characterize the damage of lithium-ion battery materials. However, there are problems in both spatial scale and time scale. In terms of spatial scale, researchers usually select a few particles as the characterization results. In reality, the active materials contained in the battery pole piece are non-uniform, that is, researchers can obtain the characterization results of the active material structure being intact and the characterization results of the active material with severe cracks in the same sample by selecting a specific observation area. Therefore, the test method for the damage degree of active materials in lithium-ion batteries should be statistically significant, and the overall damage should be tested in as many particle characterization results as possible, while the method of manual identification and statistics is not competent for the damage measurement task of hundreds or thousands of active materials. In terms of time scale, researchers usually use the method of disassembling the battery for off-site characterization to characterize the damage of active materials, which can only obtain the characterization image at a single moment, and cannot test the evolution of the damage degree of active materials under different charge and discharge cycles. Therefore, the test of the damage degree of active materials in lithium-ion batteries should have temporal continuity, and it is necessary to use a real-time observation method to observe, record and analyze the fixed area of ​​the same pole piece at different times. Summary of the invention

[0004] In response to the problems raised by the background technology, the present invention proposes a statistical method for the damage degree of active materials of lithium-ion batteries based on deep learning.

[0005] A method for testing the damage degree of active materials of lithium-ion batteries based on deep learning, comprising the following steps:

[0006] (1) Construction of intelligent identification tool for active materials in target test:

[0007] S1. Grind and polish the cross section of the lithium-ion battery electrode piece by polishing means to obtain a flat and smooth cross-section sample of the battery electrode piece;

[0008] S2. Collect cross-sectional images of lithium-ion battery electrodes as original data sets using an optical microscope;

[0009] S3. Differentiate and label the lithium-ion battery active materials of the target test in the pole piece image of the original data set, use the professional labeling tool LabelMe to draw the active materials in the cross section of the lithium-ion battery pole piece using polygons, and save the coordinate data of each vertex at the pixel scale to obtain the training data set;

[0010] S4. Construct a deep learning model Mask R-CNN, select a training data set, pre-process the training data set images and input them into the neural network for training. After gradient back propagation, calculate the loss result. If the specified cycle is not reached, continue to input the neural network for training. If the specified cycle is reached, save the neural network parameters as the target test active material intelligent identification tool;

[0011] (2) Real-time monitoring of lithium-ion battery active material damage based on in-situ optical microscope observation platform:

[0012] S5. Build an in-situ optical microscope observation platform for lithium-ion battery active materials, and monitor the cross-sectional images of lithium-ion battery pole pieces in real time during the cycle charge and discharge process;

[0013] S6. Based on the target test active material intelligent identification tool, complete the intelligent identification of the target test lithium-ion battery active material in the optical microscope image of the battery pole piece recorded in real time during the cycle charge and discharge process, and obtain the mask image and coordinate information of each active material in the image;

[0014] (3) Test and evaluate the damage degree of active materials in lithium-ion batteries:

[0015] S7. Use the threshold setting method to determine the grayscale value of the image of each active material obtained in step (2), divide the particle area into the crack region area and the block area, the crack degree of a single active material = crack region area ÷ particle area, test and analyze the crack degree of each active material to obtain the overall damage degree of the active particles of the electrode distribution to be observed.

[0016] In some embodiments, the lithium-ion battery active material is an active material particle in a battery pole piece, which may be lithium iron phosphate LiFePO 4 , LiCoO 2 、NCM ternary material LiNix Co y Mn 1-x-y O 2 , Lithium titanate Li 4 Ti 5 O 12 One of graphite C and silicon Si.

[0017] In some embodiments, the polishing method for preparing the lithium-ion battery pole piece cross-section sample in step S1 may use mechanical polishing, argon ion beam polishing or focused ion beam polishing.

[0018] In some embodiments, the in-situ optical microscope observation system in step S5 is composed of an optical microscope, a lithium-ion battery charging and discharging device, an optical in-situ reaction cell and a computer; the electrode to be observed, the counter electrode and the diaphragm in the optical in-situ reaction cell are clamped in a stacked manner to ensure that the cross-sections of the electrode to be observed, the counter electrode and the diaphragm facing the window are flush, and the side is close to the reaction cell window; the electrode to be observed is polished using an argon ion beam to obtain a flat and smooth cross-section, and this surface is used as the cross-section close to the reaction cell window.

[0019] The cyclic charge and discharge process in step S5 is: setting the optical in-situ reaction cell to charge-rest-rate discharge-rest-cycle charge and discharge steps, wherein the voltage range is 2.8-4.7V, and the charge and discharge rate is set to 0.5C.

[0020] The beneficial effect of the present invention is that the present invention discloses a method for testing the damage degree of active materials of lithium-ion batteries based on deep learning. Compared with the prior art, the improvement of the present invention is:

[0021] The present invention designs a method for testing the damage degree of active materials of lithium-ion batteries based on deep learning. The method is based on the Mask R-CNN image instance segmentation model. During the cyclic charge and discharge process of the lithium-ion battery, the image of the pole piece of the lithium-ion battery material is collected by an in-situ optical microscope method, and the Mask R-CNN model is used for prediction to realize the rapid intelligent separation of active particles in the image. At the same time, the gray value determination method using the threshold setting is used to realize the test and analysis of the damage degree. The method can perform automated and intelligent quantitative testing on the damage degree of active materials in different charge and discharge stages of the same pole piece. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of a method for testing the degree of damage of lithium-ion active battery materials based on deep learning of the present invention;

[0023] Figure 2a This is an original optical microscope image of a cross section obtained by polishing the NCM811 pole piece obtained by disassembling the soft-pack battery in the embodiment of the present application;

[0024] Figure 2b This is the LabelMe annotation result image based on the optical microscope image;

[0025] Figure 2c is the mask image corresponding to the annotated image;

[0026] Figure 3a This is a schematic diagram of the in-situ optical microscope observation platform used in the damage degree test process of lithium-ion battery active materials;

[0027] Figure 3b is a schematic diagram of an optical in-situ reaction cell;

[0028] Figure 4 This is a result diagram of the active material intelligent identification of the electrode to be observed in the embodiment of the present application using the target test active material intelligent identification tool based on the Mask R-CNN model.

[0029] Figure 5a It is a schematic diagram of active particle damage obtained based on a gray value threshold after the observed electrode piece in the embodiment of the present application performs intelligent particle identification;

[0030] Figure 5b The figure shows the test results of the damage degree of the fresh electrode and the electrode after cycling at a cut-off voltage of 4.7V. DETAILED DESCRIPTION

[0031] The present invention is further described below in conjunction with the accompanying drawings and specific implementation examples, but does not limit the protection scope of the patent of the present invention.

[0032] Implementation example: A method for testing the damage degree of active materials of lithium-ion batteries based on deep learning, specifically comprising the following steps:

[0033] (1) Construction of intelligent identification tool for active materials in target test:

[0034] Preparation of lithium-ion battery pole piece samples: A commercial NCM811 / Graphite soft-pack battery with a capacity of 4Ah was selected and disassembled in a glove box filled with argon gas (water and oxygen values ​​were all less than 0.1PPM). The obtained NCM811 pole piece was soaked in DMC solvent for 1 hour, the NCM811 pole piece was taken out and dried in a vacuum drying oven at 80°C, and the NCM811 pole piece was taken out after 12 hours.

[0035] Pre-polishing of cross-section samples of lithium-ion battery pole pieces: Cut the NCM811 pole pieces into small square pieces of 10mm*10mm. Select two 10mm*10mm silicon wafers with a thickness of 1mm, and use 502 glue to glue 10 NCM811 square pole pieces and another silicon wafer on one silicon wafer in sequence. Place the sample block in a metallographic specimen cold mounting mold, use Struers 40200029Epofix series cold mounting materials, mix the resin and curing agent at a mass ratio of 25:3, stir for 2 minutes and pour into the mold. After 12 hours, the cold mounting block is completely cured and then removed from the mold.

[0036] The cold-mounted blocks were ground and polished using a Struers Tegramin-20 automatic grinder and polisher. 400-mesh, 800-mesh, 1200-mesh, 2000-mesh and 4000-mesh sandpapers were used to grind the sample surfaces for 1 min, 1.5 min, 2 min, 2.5 min and 3 min, respectively. Subsequently, 3 μm and 1 μm diamond polishing liquids and corresponding Dac polishing cloths and Nap polishing cloths were used to rough-polish the sample surfaces for 5 min each. Finally, 0.05 μm silica polishing liquid and Chem polishing cloth were used to final-polish the sample surfaces to obtain a smooth sample surface. The cold-mounted block pressure during all polishing processes was 5 N, and the cold-mounted block speed and turntable speed were both 50 r / min and rotated in the same direction.

[0037] After final polishing, the cold mounting material of the cold mounting block except the silicon wafer clamping part is ground off to obtain a sample block with a size of about 10mm*10mm*3mm.

[0038] Polishing of lithium-ion battery pole piece cross-section samples: Use the desktop ion grinding and polishing instrument SEMPrep2 to perform argon ion beam polishing on the pre-polished surface of the sample block, so that a large number of NCM811 pole piece samples with flat and smooth cross-sections can be obtained through a single argon ion beam cutting.

[0039] Acquisition of the original data set: The cross-sectional images of NCM811 were recorded using the EC Epiplan 100x / 0.85M27 objective lens of the Zeiss Axio Imager 1 metallographic microscope. Figure 2a shown.

[0040] Acquisition of training data set: LabelMe annotation tool is used to annotate NCM811 particles in the original data set image. Using polygons, independent particles are drawn by naked eye identification and named as "particle_001", "particle_002", "particle_003"..., such as Figure 2b The annotated image is saved in .json format, and then its corresponding mask image is generated, as shown in Figure 2cAs shown, you can enter the Mask R-CNN model for subsequent training and prediction.

[0041] Construction and training of instance segmentation deep learning model: Mask R-CNN model is used to intelligently identify NCM811 active particles. The experimental environment for model construction is computer configuration: CPU: AMD EPYC 7543 32-Core Processor@3395.407MHz, GPU: NVIDIA A40 (48G), memory: 90G. The deep learning framework uses Tensorflow 1.13.1 and Keras 2.2.5. The parallel computing framework uses CUDA 11.0, and the deep neural network acceleration library is cuDNN 8.0.5.

[0042] Input the labeled training data set into the Mask R-CNN model to start training. During the training process of the Mask R-CNN model, set the batch size to 1, the number of iterations to 100, and the learning rate to 0.001. After the training is completed, the available Mask R-CNN model parameters are obtained as the target test active material intelligent identification tool.

[0043] (2) Real-time monitoring of lithium-ion battery active material damage based on in-situ optical microscope observation platform:

[0044] Preparation of the pole piece to be observed: NCM811 active material, SuperP, PVDF and NMP solution are mixed into NCM811 slurry, and the slurry is coated on the aluminum foil current collector to make the NCM811 pole piece, and the NCM811 pole piece is cut by argon ion beam to obtain a flat and smooth cross-section to be observed.

[0045] Assembly of the optical in-situ reaction cell: The optical in-situ reaction cell includes a PEEK shell, a PEEK upper cover, a quartz glass window, a sealing gasket, a sealing ring, a stainless steel conductive column, a battery material clamp and fastening screws. Use fastening screws to connect and fix the PEEK upper cover to the PEEK shell, and use this to press the quartz window glass, sealing ring and sealing gasket installed on the PEEK shell to achieve sealing. Place the prepared NCM811 electrode to be observed, PP diaphragm and counter electrode graphite electrode between two battery material clamps and clamp them, and fit the clamps to the quartz window glass, where one side of the cross-section of the electrode to be observed that has been polished by the argon ion beam fits the quartz window glass; use stainless steel conductive columns to conduct the NCM811 positive electrode and the graphite negative electrode to form a conductive path for the optical in-situ reaction cell, such as Figure 3b shown.

[0046] Construction of in-situ optical microscope observation system: Place the optical in-situ reaction cell on the stage of Zeiss Axio Imager 1 metallographic microscope, and use EC Epiplan 100x / 0.85M27 objective to focus on the NCM811 electrode to be observed; Connect the positive and negative wires of the Xinwei charging and discharging device to the positive and negative stainless steel conductive columns of the optical in-situ reaction cell respectively to prepare for the charging and discharging experiment of the optical in-situ reaction cell; Connect the computer to the optical microscope and the charging and discharging device respectively to realize the control of the optical microscope, the control of the charging and discharging device, the recording of the optical microscope image results and the recording of the charging and discharging results of the optical in-situ reaction cell, such as Figure 3a As shown; adjust the optical microscope to dark field mode and use appropriate exposure time and warm and cold color values ​​to obtain clear optical microscope images.

[0047] In-situ optical microscope observation test: The optical in-situ reaction cell is set to charge and discharge at a rate - standstill - rate discharge - standstill - cycle, where the voltage range is 2.8-4.7V and the charge and discharge rate is set to 0.5C; the optical microscope image of the cross section of the electrode to be observed is recorded in the initial state and every 10 cycles, and the active material in the electrode is intelligently identified on the electrode image obtained by the in-situ observation test in combination with the target test active material intelligent identification tool to obtain the mask image of each active material and its coordinate information, such as Figure 4 shown.

[0048] (3) Test and evaluate the damage degree of active materials in lithium-ion batteries:

[0049] For each NCM811 particle, the cracks are generally black, and the blocks are generally light white. A threshold is set to determine the gray value of each particle, and the particle area is divided into the crack area and the block area. Figure 5a As shown in the figure, the damage degree of a single active particle = crack area ÷ particle area. The damage degree of each NCM811 particle is tested and evaluated to obtain the overall damage degree of the active particles distributed in the electrode to be observed; by comparing and analyzing the crack degree of active particles with different cycle numbers, the evolution of the damage degree during the cyclic charge and discharge process can be obtained, such as Figure 5b shown.

[0050] So far, the Mask R-CNN model trained based on the cross-sectional optical microscope images of lithium-ion battery materials has been used to build an intelligent identification tool for target test active materials, and the automated intelligent identification and test evaluation of the damage degree of a large number of lithium-ion battery active material particles at a large spatial scale have been realized; at the same time, combined with the in-situ optical microscope observation test method, the test method for the evolution of the damage degree of lithium-ion battery active materials has been further expanded from the time scale, so as to obtain the evolution of the damage degree of lithium-ion battery active materials during the cyclic charge and discharge process. This method provides a reliable and advanced characterization test method for the study of the evolution mechanism of the damage degree of lithium-ion battery active materials.

[0051] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for testing the damage degree of active materials of lithium-ion batteries based on deep learning, characterized in that: The following steps are included: (1) Construction of intelligent identification tool for active materials in target test: S1. Grind and polish the cross section of the lithium-ion battery electrode piece by polishing means to obtain a flat and smooth cross-section sample of the battery electrode piece; S2. Collect cross-sectional images of lithium-ion battery electrodes as original data sets using an optical microscope; S3. Differentiate and label the lithium-ion battery active materials of the target test in the pole piece image of the original data set, use the professional labeling tool LabelMe to draw the active materials in the cross section of the lithium-ion battery pole piece using polygons, and save the coordinate data of each vertex at the pixel scale to obtain the training data set; S4. Construct a deep learning model Mask R-CNN, select a training data set, pre-process the training data set images and input them into the neural network for training. After gradient back propagation, calculate the loss result. If the specified cycle is not reached, continue to input the neural network for training. If the specified cycle is reached, save the neural network parameters as the target test active material intelligent identification tool; (2) Real-time monitoring of lithium-ion battery active material damage based on in-situ optical microscope observation platform: S5. Build an in-situ optical microscope observation platform for lithium-ion battery active materials, and monitor the cross-sectional images of lithium-ion battery pole pieces in real time during the cycle charge and discharge process; S6. Based on the target test active material intelligent identification tool, complete the intelligent identification of the target test lithium-ion battery active material in the optical microscope image of the battery pole piece recorded in real time during the cycle charge and discharge process, and obtain the mask image and coordinate information of each active material in the image; (3) Test and evaluate the damage degree of active materials in lithium-ion batteries: S7. Use the threshold setting method to determine the grayscale value of the image of each active material obtained in step (2), divide the particle area into the crack region area and the block area, the crack degree of a single active material = crack region area ÷ particle area, test and analyze the crack degree of each active material to obtain the overall damage degree of the active particles of the electrode distribution to be observed.

2. The method for testing the damage degree of active materials of lithium-ion batteries based on deep learning according to claim 1, characterized in that: The target tested lithium-ion battery active materials are active material particles in the battery pole piece, such as lithium iron phosphate LiFePO4, lithium cobalt oxide LiCoO2, NCM ternary material LiNi x Co y Mn 1-x-y O2, Lithium titanate Li4Ti5O 12 One of graphite C and silicon Si.

3. The method for testing the damage degree of active materials of lithium-ion batteries based on deep learning according to claim 1, characterized in that: The polishing method for preparing the cross-section sample of the lithium-ion battery pole piece is mechanical polishing, argon ion beam polishing or focused ion beam polishing.

4. The method for testing the damage degree of active materials of lithium-ion batteries based on deep learning according to claim 1, characterized in that: The in-situ optical microscope observation platform is composed of an optical microscope, a lithium-ion battery charging and discharging device, an optical in-situ reaction cell and a computer; the electrode to be observed, the counter electrode and the diaphragm in the optical in-situ reaction cell are clamped in a stacked manner to ensure that the cross-sections of the electrode to be observed, the counter electrode and the diaphragm facing the window are flush, and the side is closely attached to the reaction cell window; The electrode to be observed is polished using an argon ion beam to obtain a flat and smooth cross section, and this surface is used as the cross section close to the reaction cell window.

5. The method for testing the damage degree of active materials of lithium-ion batteries based on deep learning according to claim 1, characterized in that: The cyclic charge and discharge process in step S5 is: setting the optical in-situ reaction cell to charge-rest-rate discharge-rest-cycle charge and discharge steps, wherein the voltage range is 2.8-4.7V, and the charge and discharge rate is set to 0.5C.

Images