A method and device for efficient sorting based on retired lithium battery cascade utilization

Abstract

The application provides a kind of based on the high-efficiency sorting method and device of retired lithium battery cascade utilization, and relates to lithium battery technical field.The high-efficiency sorting device based on the retired lithium battery cascade utilization, including fixed box, the upper surface one end of the fixed box is provided with connecting box, the connecting box is provided with screening mechanism, the screening mechanism below is provided with collection hopper, the collection hopper is fixedly connected with connecting box, the lower end of the collection hopper is fixedly connected with guide pipe;Two screening rollers are arranged in the fixed box, and the two ends of the two screening rollers are fixedly connected with the fixed box, and a plurality of material holding drawers are uniformly arranged below the screening rollers.The interval between the two screening rollers changes from small to large, thereby completing more detailed screening of lithium batteries, which can be accurately classified and made to fall into the corresponding material holding drawers, with higher screening efficiency and higher screening accuracy.

Description

Technical Field

[0001] This invention relates to the field of lithium battery technology, specifically to a method and apparatus for efficient sorting based on the cascade utilization of retired lithium batteries. Background Technology

[0002] Lithium-ion batteries are rechargeable batteries widely used in mobile devices, electric vehicles, and energy storage systems. They store and release electrical energy by the migration of lithium ions between the positive and negative electrodes. The reuse of retired lithium-ion batteries is currently a hot research topic. Reuse refers to the process of reusing retired lithium-ion batteries, no longer suitable for their original purpose, through a series of treatments and sorting. This method can maximize the lifespan of retired lithium-ion batteries and effectively reduce their environmental impact. Efficient sorting is a crucial step in this process, and currently, most battery sorting is done manually.

[0003] Manual sorting requires a large amount of manpower, places high demands on the physical strength and energy of operators, and is labor-intensive, which can easily lead to fatigue and errors. In addition, manual sorting is inefficient and slow, which cannot meet the needs of large-scale processing, resulting in low resource utilization efficiency. Furthermore, the experience and skill levels of operators vary greatly, which may lead to low sorting accuracy and misclassification. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a highly efficient sorting method and apparatus based on the cascade utilization of retired lithium batteries, solving the problem of low sorting efficiency in manual sorting.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution: a high-efficiency sorting device based on the cascade utilization of retired lithium batteries, comprising a fixed box, a connecting box at one end of the upper surface of the fixed box, a screening mechanism inside the connecting box, a collecting hopper below the screening mechanism, the collecting hopper being fixedly connected to the connecting box, and a guide pipe fixedly connected to the lower end of the collecting hopper; two screening rollers are provided inside the fixed box, with both ends of the two screening rollers being fixedly connected to the fixed box respectively, and a plurality of material drawers are evenly arranged below the screening rollers.

[0008] Through the above technical solution, the distance between the two screening rollers gradually increases, and the lithium battery slides down from one end of the two screening rollers. Depending on the model, it falls into the corresponding material drawer through the distance between the two screening rollers, thus completing a more detailed screening of the lithium battery.

[0009] Preferably, the fixed box is fixedly connected to the connecting box, the plurality of material drawers are slidably connected to the fixed box, and each of the plurality of material drawers has a handle fixedly connected to one side, and the two screening rollers are arranged at an angle.

[0010] The above technical solution avoids confusion and mixing of different types of lithium batteries, facilitating subsequent processing and management by staff, and resulting in higher screening efficiency and accuracy.

[0011] Preferably, the screening mechanism includes a connecting plate that is slidably connected to the connecting box. Connecting rods are fixedly connected to the four corners of the upper surface of the connecting plate. The end of the connecting rod away from the connecting plate is fixedly connected to the first screening plate. A second screening plate is provided on the upper surface of the first screening plate. Both the second screening plate and the first screening plate have discharge holes.

[0012] By changing the size of the discharge holes on the second and first screen plates, the discharge size of the discharge holes can be adjusted according to the size of the lithium batteries to be screened.

[0013] Preferably, the second screening plate is slidably connected to the first screening plate, and racks are symmetrically arranged on both sides of the first screening plate. The racks are fixedly connected to the first screening plate, and gears are respectively arranged on the two racks, and the gears mesh with the racks for transmission.

[0014] Preferably, one end of the gear is fixedly connected to the output end of the first drive motor, the first drive motor is fixedly connected to the fixed block, one end of the fixed block is fixedly connected to the second screen plate, and the end of the gear away from the first drive motor is rotatably connected to the connecting column.

[0015] Through the above technical solution, the output end of the first drive motor drives the gear to rotate, the gear and rack drive each other, and the gear drives the second screen plate to slide along the direction of the sliding groove through the fixed block, so that the second screen plate slides along the first screen plate.

[0016] Preferably, the first screening plate is provided with a sliding groove, and a slider is slidably connected in the sliding groove. The slider is fixedly connected to the end of the connecting column away from the gear.

[0017] Preferably, connecting blocks are symmetrically fixedly connected to both sides of the inner wall of the connecting box, and springs are provided at the upper ends of the two connecting blocks. The two ends of the two springs are respectively fixedly connected to the connecting blocks and the lower ends of the connecting plates. Connecting seats are symmetrically fixedly connected to the lower ends of the connecting plates, and rollers are rotatably connected to the two connecting seats.

[0018] Preferably, a connecting shaft is rotatably connected through the connecting box, and two eccentric wheels are fixedly connected to the connecting shaft. The two eccentric wheels are respectively in rolling cooperation with two rollers. One end of the connecting shaft is fixedly connected to the output end of the second drive motor, and the second drive motor is fixedly connected to one side of the connecting box.

[0019] Through the above technical solution, the output end of the second drive motor drives the connecting shaft to rotate, and the connecting shaft drives several eccentric wheels on the shaft to cooperate with the rollers at the lower end of the connecting plate, thereby driving the screening mechanism to vibrate up and down to promote the screening of lithium batteries.

[0020] Preferably, a feed pipe is fixedly connected to the upper end of the connecting box, a box door is provided on one side of the connecting box, the box door is rotatably connected to the connecting box, an observation window is provided on the box door, a handle groove is provided on one side of the observation window, and the observation window is made of transparent glass.

[0021] A highly efficient sorting method based on the cascade utilization of retired lithium batteries includes the following specific steps:

[0022] S1. Start the first drive motor. The output end of the first drive motor drives the gear to rotate. The gear and rack drive the second screen plate to slide along the direction of the sliding groove through the fixed block, so that the second screen plate slides along the first screen plate. By changing the position of the second screen plate and the first screen plate, the size of the dropping hole on the second screen plate and the first screen plate can be changed. The dropping size of the dropping hole can be adjusted according to the size of the lithium battery to be screened.

[0023] S2. The retired lithium batteries to be sorted are fed into the connecting box through the feed pipe. The second drive motor is started. The output end of the second drive motor drives the connecting shaft to rotate. The connecting shaft drives several eccentric wheels on the shaft to cooperate with the rollers at the lower end of the connecting plate, thereby driving the screening mechanism to vibrate up and down to promote the screening of lithium batteries.

[0024] S3. After being screened once, the lithium batteries are collected by the hopper and fall sequentially from the feed pipe onto the screening rollers in the fixed box. The distance between the two screening rollers gradually increases, and the lithium batteries slide down from one end of the two screening rollers. Depending on the model, they fall sequentially from the distance between the two screening rollers into the corresponding material drawer, thus completing a more detailed screening of the lithium batteries.

[0025] Furthermore, for lithium batteries before sorting, sensors are used to detect the attributes of retired lithium batteries, including size, appearance uniformity and integrity, weight, energy density and battery capacity. The computing unit will transmit the data from the sensors to the controller and execute machine learning algorithms to automatically adjust the screening parameters. By adjusting the screening mechanism (3), a more efficient screening process can be achieved. Specifically, the following steps are included:

[0026] Step 1: During the screening process, visual sensors, weighing sensors, and battery testing equipment are used to obtain various attribute data of lithium batteries, including size, appearance uniformity and integrity, weight, energy density, and battery capacity.

[0027] Step 2: Preprocess the collected data and label the data according to the final screening results of the lithium batteries, indicating which lithium batteries are qualified and which are unqualified.

[0028] Step 3: Extract relevant features of the lithium battery from the sensor data, including features of size, appearance uniformity and integrity, weight, energy density and battery capacity. These features will be used as input to the machine learning algorithm.

[0029] Step 4: Select the improved regression random forest algorithm to predict the quality of lithium batteries and suitable screening parameters based on sensor data;

[0030] Suppose we have a dataset that includes lithium battery dimensions, appearance uniformity and integrity, weight, energy density, and battery capacity features, as well as a quality label for each battery. We have already trained a regression model and a random forest model, and we obtain their prediction results respectively:

[0031] Suppose the prediction result of the random forest model is y^random forest; the prediction result of the regression model is y^linear regression. The former represents the screening of the internal quality of the battery, and the latter represents the estimation of the external quality of the lithium battery. The outputs of the two models are used as inputs using the stacking method, and the linear regression model is used to combine them. The final prediction result is y^final.

[0032] Each model is assigned weights w1 and w2, with values ​​between 0 and 1. Since energy density and battery capacity have a greater impact on the subsequent sorting process, these factors are prioritized for optimization in the random forest model, and w1 is correspondingly increased. The final prediction result is then obtained through a weighted average, expressed as:

[0033] y^final=w1·y^random forest+w2·y^linear regression

[0034] Optimization was performed using a random forest model with an improved structure:

[0035] Model selection and training: The Random Forest classifier is imported from the Scikit-Learn library using the Random Forest Classifier class. A random forest model with n decision trees is created using Random Forest Classifier(n_estimators=n), where the n_estimators parameter specifies the number of trees in the random forest. Different parameter values ​​can be set as needed. The random forest model is trained by fitting the training data X_train and the label y_train using rf_model.fit(X_train, y_train). rf_model is an instance of a random forest model, and fit is a method used in machine learning to fit (train) a model.

[0036] Extracting the structure of the first tree: Import from the tree module of Scikit-Learn using the export_text function. This function allows you to extract the structural information of a single decision tree. export_text(rf_model.estimators_[0], feature_names=feature_names) extracts the structural information of the first tree in the random forest. rf_model.estimators_[0] represents the first decision tree in the random forest, and feature_names is a list of feature names. export_text is a function used to convert the decision tree structure into text format.

[0037] Print the tree structure: Finally, use print(tree_structure) to print the structure information of the first decision tree to the screen as text so that you can see the tree’s splitting rules and node structure;

[0038] The structure of a decision tree:

[0039] feature_0 and feature_1 are two features used to describe the energy density and capacity attributes of the battery.

[0040] Each node in the decision tree represents a feature condition, comparing the corresponding feature values. Here, there are two feature conditions: feature_0 and feature_1.

[0041] feature_0 represents the energy density of the battery. The value ranges from 0 to 1, reflecting the level of energy density. 0 indicates a low energy density battery, and 1 indicates a high energy density battery.

[0042] feature_1 represents the battery capacity, with a value ranging from 0 to 5, reflecting the size of the battery capacity, where 0 represents a small capacity battery and 5 represents a large capacity battery;

[0043] At this point, y^random forest = q1*feature_0 + q2*feature_1, where q1 and q2 are weights of 0-1, which can be set as needed;

[0044] feature_0 <= 0.5 means a battery with low energy density;

[0045] feature_0 > 0.5 means a battery with high energy density;

[0046] feature_1 <= 1.5 means a low-capacity battery;

[0047] 1.5 < feature_1 <= 2.5 means a battery with medium capacity;

[0048] feature_1 > 2.5 indicates a large-capacity battery;

[0049] The root node of the tree starts from feature_0. It checks if it is less than or equal to 0.5. If it is, it moves to the left subtree; otherwise, it moves to the right subtree.

[0050] The left subtree further checks whether feature_1 is less than or equal to 1.5. If so, it predicts the class [1.0, 0.0], which means that under this condition, the model has a higher probability of classifying the data point as class 1.

[0051] If feature_1 in the left subtree is greater than 1.5 and less than or equal to 2.5, then it predicts the class [0.0, 1.0], indicating that under this condition, the model has a higher probability of classifying the data point as class 2;

[0052] Finally, if feature_1 in the right subtree is greater than 2.5, then it predicts the class [0.0, 1.0], which means that under this condition, the model has a higher probability of classifying the data point as class 2.

[0053] This decision tree structure can be used for classification. By following the branching conditions of the tree from the root node to a leaf node, the category of a data point can be determined. In this invention, the categories are binary: [1.0, 0.0] indicates that category 1 has a higher probability, and [0.0, 1.0] indicates that category 2 has a higher probability. Category 1 represents "poor quality", and category 2 represents "good quality". The structure information of the first decision tree is printed to the screen as text using print(tree_structure) so that the splitting rules and node structure of the tree can be viewed.

[0054] The prediction of a single decision tree is explained by the structure and rules of the tree. Each tree consists of a series of branch nodes and leaf nodes. Each branch node contains a feature and a condition rule. When a data point passes through the tree, it moves through the branches of the tree according to the condition rule until it reaches a leaf node, which contains the predicted value.

[0055] Model Assumptions: The basic assumption of the linear regression model is that the predicted outcome is a linear combination of features, that is:

[0056] y^linear regression = θ0 + θ1·size + θ2·battery surface integrity + θ3·weight + θ4·battery surface uniformity, where θ0 is the intercept term, and θ0, θ1, θ2, θ3, and θ4 are the weights of the features, which are determined by the weights of the feature importance.

[0057] Loss function: Define a loss function, usually the mean squared error, to measure the difference between the model's predictions and the true values. Where n is the number of samples, y^i is the model's predicted value, and yi is the corresponding true value;

[0058] Minimizing the loss function: By minimizing the loss function, we find the optimal parameters θ0, θ1, θ2, θ3, θ4 to minimize the loss function. This is typically accomplished using optimization algorithms such as gradient descent.

[0059] Final prediction: Once the optimal parameters are obtained, they can be used to make predictions, resulting in y^linear regression; This model judges the quality of the battery by visually detecting whether the battery is damaged and its weight.

[0060] Finally, based on the formula y^final=w1·y^linear regression+w2·y^random forest, the battery quality is sorted by combining the regression algorithm and the random forest algorithm. High-quality batteries will be sorted according to specifications, while substandard batteries will be further recycled and processed.

[0061] Step 5: Train the selected machine learning model using the labeled dataset to enable it to automatically infer the relationship between lithium battery properties and screening parameters. During the screening process, acquire lithium battery data in real time through sensors and input it into the trained machine learning model to obtain real-time predictions about lithium battery quality and screening parameters. Based on the real-time prediction results of the machine learning model, adjust the screening parameters, including sieve size, vibration frequency and amplitude, to achieve more accurate screening.

[0062] (III) Beneficial Effects

[0063] This invention provides a method and apparatus for efficient sorting of retired lithium batteries based on their cascade utilization.

[0064] It has the following beneficial effects:

[0065] 1. This invention uses gear and rack transmission. The gear drives the second screen plate to slide along the direction of the sliding groove through the fixed block, so that the second screen plate slides along the first screen plate. By changing the position of the overlap between the second screen plate and the first screen plate, the size of the dropping hole on the second screen plate and the first screen plate can be changed. The operator can adjust the dropping size of the dropping hole according to the size of the lithium battery to be screened.

[0066] 2. This invention uses a gradually increasing gap between two screening rollers. Lithium batteries slide down from one end of the two screening rollers and, according to their different models, fall sequentially into the corresponding material drawers through the gap between the two screening rollers. This achieves more detailed screening of lithium batteries, allowing them to be accurately classified and placed into the appropriate material drawers. This avoids confusion and mixing of different models of lithium batteries, facilitating subsequent processing and management by staff, resulting in higher screening efficiency and accuracy.

[0067] 3. The regression random forest algorithm employed in this invention, through an intelligent control system, allows sensors to monitor battery properties in real time, and the computing unit to automatically adjust screening parameters to ensure that only qualified batteries are retained, thereby improving the quality and efficiency of the screening process. The introduction of the intelligent system reduces the need for manual intervention, lowers the risk of human error, and improves consistency and accuracy. For unqualified batteries, appropriate recycling and disposal measures can be taken, helping to reduce environmental pollution and resource waste. The automated screening process saves manpower and time costs and improves resource utilization efficiency. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of the internal structure of the present invention;

[0069] Figure 2 This is a schematic diagram of the screening roller structure of the present invention;

[0070] Figure 3 This is a first-view structural diagram of the screening mechanism of the present invention;

[0071] Figure 4 This is a schematic diagram of the second-view structure of the screening mechanism of the present invention;

[0072] Figure 5 This is a schematic diagram of the screening mechanism of the present invention from a third-view perspective;

[0073] Figure 6 This is a cross-sectional view of the fixing box of the present invention;

[0074] Figure 7 This is a schematic diagram of the structure of the present invention.

[0075] The components are as follows: 1. Fixed box; 11. Material drawer; 12. Handle; 13. Screening roller; 2. Connecting box; 21. Feed pipe; 22. Box door; 23. Observation window; 24. Handle groove; 3. Screening mechanism; 31. Collection hopper; 32. Guide pipe; 4. Connecting plate; 41. Connecting rod; 42. First screening plate; 43. Discharge hole; 44. Second screening plate; 45. Fixed block; 46. Rack; 47. Sliding groove; 48. Sliding block; 5. Gear; 51. Connecting column; 52. First drive motor; 6. Connecting block; 61. Spring; 7. Connecting shaft; 71. Eccentric wheel; 72. Second drive motor; 73. Connecting seat; 74. Roller. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Example 1:

[0078] like Figure 1 and Figure 2This invention provides a high-efficiency sorting device for the cascade utilization of retired lithium batteries, comprising a fixed box 1, a connecting box 2 at one end of the upper surface of the fixed box 1, a screening mechanism 3 inside the connecting box 2, a collecting hopper 31 below the screening mechanism 3, the collecting hopper 31 being fixedly connected to the connecting box 2, and a guide pipe 32 fixedly connected to the lower end of the collecting hopper 31; two screening rollers 13 are arranged inside the fixed box 1, with both ends of the two screening rollers 13 being fixedly connected to the fixed box 1, and a plurality of material drawers 11 evenly arranged below the screening rollers 13, with the material drawers between the two screening rollers 13 being... The spacing between the two screening rollers 13 gradually increases, allowing the lithium batteries to slide down from one end of each roller. Depending on their model, the batteries fall sequentially into the corresponding storage drawers 11 through the gap between the rollers 13, thus achieving more precise screening of the lithium batteries. The fixed box 1 is fixedly connected to the connecting box 2, and several storage drawers 11 are slidably connected to the fixed box 1. Each storage drawer 11 has a handle 12 fixedly connected to one side. The two screening rollers 13 are set at an angle to avoid confusion and mixing of different models of lithium batteries, facilitating subsequent processing and management by staff, resulting in higher screening efficiency and accuracy.

[0079] like Figure 3 , Figure 4 and Figure 5 The screening mechanism 3 includes a connecting plate 4 slidably connected to the connecting box 2. Connecting rods 41 are fixedly connected to the four corners of the upper surface of the connecting plate 4. The end of the connecting rod 41 away from the connecting plate 4 is fixedly connected to the first screening plate 42. A second screening plate 44 is provided on the upper surface of the first screening plate 42. Both the second screening plate 44 and the first screening plate 42 have discharge holes 43. By changing the size of the discharge holes 43 on the second screening plate 44 and the first screening plate 42, the discharge size of the discharge holes 43 can be adjusted according to the required lithium battery specifications. The second screening plate 44 is slidably connected to the first screening plate 42. Racks 46 are symmetrically arranged on both sides of the first screening plate 42 and are fixedly connected to the first screening plate 42. Each of the two racks 46 has a... A gear 5 is provided, which meshes with a rack 46 for transmission. One end of the gear 5 is fixedly connected to the output end of a first drive motor 52. The first drive motor 52 is fixedly connected to a fixed block 45. One end of the fixed block 45 is fixedly connected to a second screen plate 44. The end of the gear 5 away from the first drive motor 52 is rotatably connected to a connecting column 51. The output end of the first drive motor 52 drives the gear 5 to rotate. The gear 5 and the rack 46 are driven to transmit power. The gear 5 drives the second screen plate 44 to slide along the direction of the sliding groove 47 through the fixed block 45, so that the second screen plate 44 slides along the first screen plate 42. A sliding groove 47 is provided on the first screen plate 42. A slider 48 is slidably connected in the sliding groove 47. The slider 48 is fixedly connected to the end of the connecting column 51 away from the gear 5.

[0080] like Figure 6 and Figure 7As shown, connecting blocks 6 are symmetrically fixedly connected to both sides of the inner wall of the connecting box 2. A spring 61 is provided at the upper end of each connecting block 6. The two ends of the two springs 61 are respectively fixedly connected to the connecting block 6 and the lower end of the connecting plate 4. Connecting seats 73 are symmetrically fixedly connected to the lower end of the connecting plate 4. Rollers 74 are rotatably connected to the two connecting seats 73. A connecting shaft 7 is rotatably connected through the connecting box 2. Two eccentric wheels 71 are fixedly connected to the connecting shaft 7. The two eccentric wheels 71 roll in cooperation with the two rollers 74. One end of the connecting shaft 7 is fixedly connected to the output end of the second drive motor 72. The second drive motor 72 is fixedly connected to one side of the connecting box 2. The output end of the second drive motor 72 drives the connecting shaft 7 to rotate. The connecting shaft 7 drives several eccentric wheels 71 on the shaft to cooperate with the rollers 74 at the lower end of the connecting plate 4, thereby driving the screening mechanism 3 to vibrate up and down to promote the screening of lithium batteries. The upper end of the connecting box 2 is fixedly connected to the feed pipe 21. A box door 22 is provided on one side of the connecting box 2. The box door 22 is rotatably connected to the connecting box 2. An observation window 23 is provided on the box door 22. A handle groove 24 is opened on one side of the observation window 23. The observation window 23 is made of transparent glass.

[0081] A highly efficient sorting method based on the cascade utilization of retired lithium batteries includes the following specific steps:

[0082] S1. Start the first drive motor 52. The output end of the first drive motor 52 drives the gear 5 to rotate. The gear 5 is driven by the rack 46. The gear 5 drives the second screen plate 44 to slide along the direction of the sliding groove 47 through the fixed block 45, so that the second screen plate 44 slides along the first screen plate 42. By changing the position of the second screen plate 44 and the first screen plate 42, the size of the dropping hole 43 on the second screen plate 44 and the first screen plate 42 can be changed. The dropping size of the dropping hole 43 can be adjusted according to the size of the lithium battery to be screened.

[0083] S2. The retired lithium batteries to be sorted are fed into the connecting box 2 through the feed pipe 21. The second drive motor 72 is started. The output end of the second drive motor 72 drives the connecting shaft 7 to rotate. The connecting shaft 7 drives several eccentric wheels 71 on the shaft to cooperate with the rollers 74 at the lower end of the connecting plate 4, thereby driving the screening mechanism 3 to vibrate up and down to promote the screening of lithium batteries.

[0084] S3. After being screened once, the lithium batteries are collected by the collection hopper 31 and fall sequentially from the guide pipe 32 onto the screening rollers 13 in the fixed box 1. The distance between the two screening rollers 13 gradually increases. The lithium batteries slide down from one end of the two screening rollers 13 and, depending on their model, fall sequentially from the distance between the two screening rollers 13 into the corresponding material drawer 11, thus completing a more detailed screening of the lithium batteries.

[0085] For lithium batteries before sorting, sensors are used to detect the attributes of retired lithium batteries, including size, appearance uniformity and integrity, weight, energy density and battery capacity. The computing unit will transmit the data from the sensors to the controller and execute machine learning algorithms to automatically adjust the screening parameters. By adjusting the screening mechanism (3), a more efficient screening process can be achieved. Specifically, the following steps are included:

[0086] Step 1: During the screening process, visual sensors, weighing sensors, and battery testing equipment are used to obtain various attribute data of lithium batteries, including size, appearance uniformity and integrity, weight, energy density, and battery capacity.

[0087] Step 2: Preprocess the collected data and label the data according to the final screening results of the lithium batteries, indicating which lithium batteries are qualified and which are unqualified.

[0088] Step 3: Extract relevant features of the lithium battery from the sensor data, including features of size, appearance uniformity and integrity, weight, energy density and battery capacity. These features will be used as input to the machine learning algorithm.

[0089] Step 4: Select the improved regression random forest algorithm to predict the quality of lithium batteries and suitable screening parameters based on sensor data;

[0090] Suppose we have a dataset that includes lithium battery dimensions, appearance uniformity and integrity, weight, energy density, and battery capacity features, as well as a quality label for each battery. We have already trained a regression model and a random forest model, and we obtain their prediction results respectively:

[0091] Suppose the prediction result of the random forest model is y^random forest; the prediction result of the regression model is y^linear regression. The former represents the screening of the internal quality of the battery, and the latter represents the estimation of the external quality of the lithium battery. The outputs of the two models are used as inputs using the stacking method, and the linear regression model is used to combine them. The final prediction result is y^final.

[0092] Each model is assigned weights w1 and w2, with values ​​between 0 and 1. Since energy density and battery capacity have a greater impact on the subsequent sorting process, these factors are prioritized for optimization in the random forest model, and w1 is correspondingly increased. The final prediction result is then obtained through a weighted average, expressed as:

[0093] y^final=w1·y^random forest+w2·y^linear regression

[0094] Optimization was performed using a random forest model with an improved structure:

[0095] Model selection and training: The Random Forest classifier is imported from the Scikit-Learn library using the Random Forest Classifier class. A random forest model with n decision trees is created using Random Forest Classifier(n_estimators=n), where the n_estimators parameter specifies the number of trees in the random forest. Different parameter values ​​can be set as needed. The random forest model is trained by fitting the training data X_train and the label y_train using rf_model.fit(X_train, y_train). rf_model is an instance of a random forest model, and fit is a method used in machine learning to fit (train) a model.

[0096] Extracting the structure of the first tree: Import from the tree module of Scikit-Learn using the export_text function. This function allows you to extract the structural information of a single decision tree. export_text(rf_model.estimators_[0], feature_names=feature_names) extracts the structural information of the first tree in the random forest. rf_model.estimators_[0] represents the first decision tree in the random forest, and feature_names is a list of feature names. export_text is a function used to convert the decision tree structure into text format.

[0097] Print the tree structure: Finally, use print(tree_structure) to print the structure information of the first decision tree to the screen as text so that you can see the tree’s splitting rules and node structure;

[0098] The structure of a decision tree:

[0099] feature_0 and feature_1 are two features used to describe the energy density and capacity attributes of the battery.

[0100] Each node in the decision tree represents a feature condition, comparing the corresponding feature values. Here, there are two feature conditions: feature_0 and feature_1.

[0101] feature_0 represents the energy density of the battery. The value ranges from 0 to 1, reflecting the level of energy density. 0 indicates a low energy density battery, and 1 indicates a high energy density battery.

[0102] feature_1 represents the battery capacity, with a value ranging from 0 to 5, reflecting the size of the battery capacity, where 0 represents a small capacity battery and 5 represents a large capacity battery;

[0103] At this point, y^random forest = q1*feature_0 + q2*feature_1, where q1 and q2 are weights of 0-1, which can be set as needed;

[0104] feature_0 <= 0.5 means a battery with low energy density;

[0105] feature_0 > 0.5 means a battery with high energy density;

[0106] feature_1 <= 1.5 means a low-capacity battery;

[0107] 1.5 < feature_1 <= 2.5 means a battery with medium capacity;

[0108] feature_1 > 2.5 indicates a large-capacity battery;

[0109] The root node of the tree starts from feature_0. It checks if it is less than or equal to 0.5. If it is, it moves to the left subtree; otherwise, it moves to the right subtree.

[0110] The left subtree further checks whether feature_1 is less than or equal to 1.5. If so, it predicts the class [1.0, 0.0], which means that under this condition, the model has a higher probability of classifying the data point as class 1.

[0111] If feature_1 in the left subtree is greater than 1.5 and less than or equal to 2.5, then it predicts the class [0.0, 1.0], indicating that under this condition, the model has a higher probability of classifying the data point as class 2;

[0112] Finally, if feature_1 in the right subtree is greater than 2.5, then it predicts the class [0.0, 1.0], which means that under this condition, the model has a higher probability of classifying the data point as class 2.

[0113] This decision tree structure can be used for classification. By following the branching conditions of the tree from the root node to a leaf node, it is possible to determine which category a data point belongs to. In this invention, the categories are binary: [1.0, 0.0] indicates that category 1 has a higher probability, and [0.0, 1.0] indicates that category 2 has a higher probability. Category 1 represents "poor quality", and category 2 represents "good quality". Scoring can be assigned during the calculation process, and the specific values ​​are selected according to the actual situation.

[0114] Use print(tree_structure) to print the structure information of the first decision tree to the screen as text so that you can see the tree’s splitting rules and node structure;

[0115] The prediction of a single decision tree is explained by the structure and rules of the tree. Each tree consists of a series of branch nodes and leaf nodes. Each branch node contains a feature and a condition rule. When a data point passes through the tree, it moves through the branches of the tree according to the condition rule until it reaches a leaf node, which contains the predicted value.

[0116] Model Assumptions: The basic assumption of the linear regression model is that the predicted outcome is a linear combination of features, that is:

[0117] y^linear regression = θ0 + θ1·size + θ2·battery surface integrity + θ3·weight + θ4·battery surface uniformity, where θ0 is the intercept term, and θ0, θ1, θ2, θ3, and θ4 are the weights of the features, which are determined by the weights of the feature importance.

[0118] Loss function: Define a loss function, usually the mean squared error, to measure the difference between the model's predictions and the true values. Where n is the number of samples, y^i is the model's predicted value, and yi is the corresponding true value;

[0119] Minimizing the loss function: By minimizing the loss function, we find the optimal parameters θ0, θ1, θ2, θ3, θ4 to minimize the loss function. This is typically accomplished using optimization algorithms such as gradient descent.

[0120] Final prediction: Once the optimal parameters are obtained, they can be used to make predictions, resulting in y^linear regression; the values ​​obtained can be set according to the actual situation. This model judges the quality of the battery by visually detecting whether the battery is damaged and its weight.

[0121] Finally, based on the formula y^final=w1·y^linear regression+w2·y^random forest, the battery quality is sorted by the regression random forest algorithm. High-quality batteries will be sorted according to specifications, while substandard batteries will be further recycled and processed.

[0122] Step 5: Train the selected machine learning model using the labeled dataset to enable it to automatically infer the relationship between lithium battery properties and screening parameters. During the screening process, acquire lithium battery data in real time through sensors and input it into the trained machine learning model to obtain real-time predictions about lithium battery quality and screening parameters. Based on the real-time prediction results of the machine learning model, adjust the screening parameters, including sieve size, vibration frequency and amplitude, to achieve more accurate screening.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-efficiency sorting method based on the cascade utilization of retired lithium batteries, comprising a fixed box, characterized in that: The upper end of the fixed box is provided with a connecting box, a screening mechanism is arranged in the connecting box, a collecting hopper is arranged below the screening mechanism, the collecting hopper is fixedly connected with the connecting box, and a guide pipe is fixedly connected to the lower end of the collecting hopper; Two screening rollers are arranged in the fixed box, and the two ends of the two screening rollers are fixedly connected with the fixed box; a plurality of material containing drawers are uniformly arranged below the screening rollers; The screening mechanism comprises a connecting plate which is slidably connected with the connecting box, connecting rods are fixedly connected to the upper surface of the connecting plate at four corner positions, respectively, a first screening plate is fixedly connected to the end of the connecting rod away from the connecting plate, a second screening plate is arranged on the upper surface of the first screening plate, and a material falling hole is formed in the upper surface of the first screening plate and the second screening plate; The second screening plate is slidably connected with the first screening plate, a rack is symmetrically arranged on the two sides of the first screening plate, the rack is fixedly connected with the first screening plate, a gear is arranged on each of the two racks, and the gears are in meshing transmission with the racks; One end of the gear is fixedly connected with the output end of a first driving motor, the first driving motor is fixedly connected with a fixed block, one end of the fixed block is fixedly connected with the second screening plate, and the end of the gear away from the first driving motor is rotatably connected with a connecting column; A sliding groove is formed in the first screening plate, a sliding block is slidably connected in the sliding groove, the sliding block is fixedly connected with the end of the connecting column away from the gear, connecting seats are symmetrically fixedly connected to the lower end of the connecting plate, and rollers are rotatably connected to the two connecting seats; A connecting shaft is rotatably connected in the connecting box, two eccentric wheels are fixedly connected to the connecting shaft, the two eccentric wheels are in rolling cooperation with each other and the two rollers, respectively, one end of the connecting shaft is fixedly connected with the output end of a second driving motor, the second driving motor is fixedly connected with one side of the connecting box, and a feeding pipe is fixedly connected to the upper end of the connecting box; The efficient sorting method for the step-by-step utilization of retired lithium batteries comprises the following specific steps: S1, start the first driving motor, the output end of the first driving motor drives the gear to rotate, the gear drives the rack in transmission, the gear drives the second screening plate to slide along the direction of the sliding groove through the fixed block, so that the second screening plate slides along the first screening plate, the size of the material falling hole of the second screening plate and the first screening plate is changed by changing the position of the second screening plate coinciding with the first screening plate, and the size of the material falling hole can be adjusted according to the required size of the lithium battery to be screened; S2, put the retired lithium batteries to be sorted into the connecting box from the feeding pipe, start the second driving motor, the output end of the second driving motor drives the connecting shaft to rotate, the connecting shaft drives the eccentric wheels on the shaft to cooperate with the rollers at the lower end of the connecting plate, so as to drive the screening mechanism to vibrate up and down and promote the screening of the lithium batteries; S3, the lithium batteries after one screening are collected by the collecting hopper and then fall into the screening rollers in the fixed box from the guide pipe one by one, the distance between the two screening rollers changes from small to large, the lithium batteries slide down from one end of the two screening rollers, and according to different models, the lithium batteries fall into the corresponding material containing drawers from the distance between the two screening rollers, so that the screening of the lithium batteries is more detailed. For the sorting of lithium batteries before sorting, the attributes of the retired lithium batteries are detected by sensors and detection equipment, and the calculation unit will transmit the sensor data to the controller to execute a machine learning algorithm to automatically adjust the screening parameters, and the screening mechanism is adjusted to realize a more efficient screening process; Specifically, the following steps are included: Step 1, in the screening process, use visual sensors, weighing sensors, and battery detection equipment to obtain various attribute data of lithium batteries, including size, appearance uniformity and integrity, weight, energy density and battery capacity; Step 2, pre-process the collected data, and label the data according to the final screening results of the lithium batteries, indicating which lithium batteries are qualified and which are not; Step 3, extract relevant features of lithium batteries from sensor data, including size, appearance uniformity and integrity, weight, energy density and battery capacity, which will be used as input for the machine learning algorithm; Step 4, select an improved regression random forest algorithm to predict the quality of lithium batteries and suitable screening parameters based on sensor data; Assuming there is a dataset including size, appearance uniformity and integrity, weight, energy density and battery capacity features of lithium batteries, and quality labels for each battery, a regression model and a random forest model have been trained, and their prediction results are obtained respectively: Assuming the prediction result of the random forest model is y^random forest; the prediction result of the regression model is y^linear regression, the former represents the prediction of the internal quality of the battery, and the latter represents the estimation of the external quality of the lithium battery; Use the stacking method to use the outputs of the two models as input, and use the linear regression model to combine them, and the final prediction result is y^final; Assign weights w1 and w2 to each model, w1 and w2 are values between 0 and 1, since energy density and battery capacity factors have a greater impact on the sorting process, energy density and battery capacity are optimized in the random forest model, w1 is assigned accordingly. Higher value, then the final prediction result is obtained by weighted average, represented as: y^final=w1·y^random*forest+w2·y^linear regression; Use the labeled dataset to train the selected machine learning model to automatically infer the relationship between lithium battery attributes and screening parameters, and in the screening process, real-time data of lithium batteries are obtained through sensors and input into the trained machine learning model to obtain real-time predictions about lithium battery quality and screening parameters. According to the real-time prediction results of the machine learning model, adjust the screening parameters, including screen size, vibration frequency and amplitude, to achieve more accurate screening.

2. The efficient sorting method based on the cascade utilization of retired lithium batteries according to claim 1, characterized in that: The fixed box is fixedly connected with the connecting box, the plurality of material containing drawers are slidably connected with the fixed box, a handle is fixedly connected to one side of each of the plurality of material containing drawers, and the two screening rollers are inclined.

3. The efficient sorting method based on the step-by-step utilization of retired lithium batteries according to claim 1, characterized in that: The connecting box is provided with connecting blocks symmetrically fixed on the inner wall of the connecting box, springs are arranged on the upper ends of the two connecting blocks, and the two ends of the two springs are fixedly connected with the connecting blocks and the lower end of the connecting plate.

4. The efficient sorting method based on the step-by-step utilization of retired lithium batteries according to claim 1, characterized in that: One side of the connecting box is provided with a box door, the box door is rotationally connected with the connecting box, an observation window is arranged on the box door, a handle slot is formed on one side of the observation window, and the observation window is made of transparent glass.

5. The efficient sorting method based on the gradient utilization of retired lithium batteries according to claim 1, characterized in that, The random forest model with the improved structure is used for optimization: Model selection and training: Import the random forest classifier from the Scikit-Learn library using the Random Forest Classifier class, create a random forest model containing n decision trees by Random Forest Classifier(n_estimators=n), where the n_estimators parameter specifies the number of trees in the random forest, and set different parameter values as needed; use rf_model.fit(X_train,y_train) to fit the training data X_train and labels y_train, and train the random forest model; rf_model is an instance of the random forest model, and fit is a method used to fit the training model in machine learning; Extract the structure of the first tree: Import the export_text function from the tree module of Scikit-Learn, which allows you to extract the structure information of a single decision tree, export_text(rf_model.estimators_[0],feature_names=feature_names) extracts the structure information of the first tree in the random forest, rf_model.estimators_[0] represents the first decision tree in the random forest, and feature_names is a list of feature names; export_text is a function for converting the structure of a decision tree into text format; Print the structure of the tree: Finally, use print(tree_structure) to print the structure information of the first decision tree in text form to the screen, so that you can view the splitting rules and node structure of the tree; The structure of the decision tree: feature_0 and feature_1 are two features that describe the energy density and battery capacity attributes of the battery; Each node of the decision tree represents a feature condition, which compares the corresponding feature value. Here, there are two feature conditions: feature_0 and feature_1. feature_0 represents the energy density of the battery, with a value range of 0 to 1, reflecting the degree of energy density, where 0 represents a small energy density battery and 1 represents a large energy density battery. feature_1 represents the battery capacity of the battery, with a value range of 0 to 5, reflecting the degree of battery capacity, where 0 represents a small capacity battery and 5 represents a large capacity battery. At this time, y^random forest = q1 * feature_0 + q2 * feature_1, where q1 and q2 are weights between 0 and 1, set as needed; feature_0 <= 0.5 means a small energy density battery; feature_0 > 0.5 means a large energy density battery; feature_1 <= 1.5 means a low battery capacity battery; 1.5 < feature_1 <= 2.5 means a medium battery capacity battery; feature_1 > 2.5 means a large battery capacity battery; The root node of the tree starts with feature_0, checking if it is less than or equal to 0.5, if so, it turns to the left subtree, otherwise it turns to the right subtree; The left subtree further checks if feature_1 is less than or equal to 1.5, if so, it predicts the class [1.0, 0.0], which means that under this condition, the model has a higher probability of classifying the data point as class 1; If feature_1 in the left subtree is greater than 1.5 and less than or equal to 2.5, it predicts the class [0.0, 1.0], indicating that under this condition, the model has a higher probability of classifying the data point as class 2; Finally, if feature_1 in the right subtree is greater than 2.5, it predicts the class [0.0, 1.0], indicating that under this condition, the model has a higher probability of classifying the data point as class 2; The structure of this decision tree can be used for classification, by starting from the root node and following the branch conditions of the tree, eventually reaching a leaf node, it can be determined which category the data point belongs to, the category is binary, [1.0, 0.0] means the probability of class 1 is higher, [0.0, 1.0] means the probability of class 2 is higher; class 1 represents "poor quality", class 2 represents "good quality"; use print(tree_structure) to print the structure information of the first decision tree to the screen in text form, so that you can view the splitting rules and node structure of the tree; The prediction of a single decision tree is explained by the structure and rules of the tree, each tree is composed of a series of branch nodes and leaf nodes, each branch node contains a feature and a conditional rule, when the data point passes through the tree, it will advance in the branches of the tree according to the conditional rule, until it reaches a leaf node, the leaf node contains the prediction value; Model assumptions: the basic assumption of the linear regression model is that the predicted result is a linear combination of features, i.e.: y^linear regression = θ0 + θ1 * size + θ2 * battery surface integrity + θ3 * weight + θ4 * battery surface uniformity, where θ0 is the intercept term, θ0, θ1, θ2, θ3, θ4 are the weights of the features, determined by the feature importance weight; Loss Function: Define a loss function, usually Mean Squared Error (MSE), to measure the difference between the model's predicted values and the true values: where n is the number of samples, y^i is the model's predicted value, and yi is the corresponding true value. Minimize the loss function: by minimizing the loss function, find the optimal parameters θ0, θ1, θ2, θ3, θ4 to minimize the loss function; this is usually done using optimization algorithms such as gradient descent; Final prediction: Once the optimal parameters are obtained, these parameters can be used to make predictions, resulting in y^linearregression; this model can detect the integrity of the battery through visual inspection and determine the quality of the battery through weight; Finally, according to the formula: y^final=w1·y^random forest+w2·y^linear regression, the battery quality is sorted through regression algorithm and random forest algorithm. Good quality batteries will be sorted according to specifications, and batteries that do not meet quality standards will be further recycled and processed.

Images