A Smart Sorting and Targeted Dismantling Method for Lithium-ion Battery Recycling

Abstract

This application provides an intelligent sorting and targeted dismantling method for lithium battery recycling. By applying a specific frequency pulse voltage to waste lithium batteries, electrochemical response data is collected. Impedance spectrum features are extracted using a convolutional neural network, and a support vector machine classifier is used to classify the health status into multiple categories, generating classification labels. Then, based on remaining capacity and cycle life estimates, the sorting batch sequence is optimized, and finally, control commands are sent to automated conveying equipment to achieve automated sorting. This application significantly improves classification accuracy and processing efficiency, and reduces the risk of misjudgment.

Description

Technical Field

[0001] This invention relates to the field of lithium battery recycling technology, and in particular to an intelligent sorting and directional dismantling method for lithium battery recycling. Background Technology

[0002] The lithium battery recycling sector plays an indispensable role in promoting resource recycling and environmental protection. With the rapid development of the new energy industry, the number of used lithium batteries has increased dramatically. How to efficiently and accurately process these batteries is not only crucial to the efficiency of resource reuse but also directly impacts sustainable environmental development. However, this field currently faces numerous challenges and urgently requires innovative technologies to overcome these bottlenecks.

[0003] Currently, the methods for recycling and processing used lithium batteries have significant shortcomings in practical applications. Many traditional methods rely on manual judgment or simple mechanical sorting, which is insufficient to handle the diverse types and complex conditions of batteries. This crude approach not only leads to inaccurate sorting results but also significantly reduces the efficiency of subsequent resource utilization, and may even result in the misprocessing of some reusable batteries, wasting valuable resources.

[0004] A deeper issue lies in the fact that assessing the health status of spent lithium batteries has become a core technical challenge. The health status of a battery directly determines whether it is suitable for reuse or needs to be dismantled and recycled, and this assessment relies on precise analysis of the battery's internal characteristics. In particular, the electrochemical properties of batteries vary significantly under different usage scenarios. For example, some batteries may have undergone intensive use, resulting in severe internal damage, while others may only be mildly aged and still have considerable residual value. If these differences in characteristics cannot be accurately captured, it becomes difficult to determine a suitable processing path for each battery, thus affecting the efficiency of the entire recycling process.

[0005] Therefore, in the complex scenario of recycling waste lithium batteries, accurately identifying the electrochemical characteristics of each battery and formulating reasonable classification and treatment strategies based on these characteristics has become a key issue in improving recycling efficiency and resource utilization. Summary of the Invention

[0006] To address the technical problems mentioned in the background section, this invention provides an intelligent sorting and directional dismantling method for lithium battery recycling, comprising:

[0007] S1, apply a pulse voltage stimulus of a specific frequency to the waste lithium battery to obtain the initial response current and voltage change data generated by the waste lithium battery;

[0008] S2, feature extraction is performed on the initial response current and voltage change data to obtain the electrochemical impedance spectroscopy feature vector;

[0009] S3, determine whether the electrochemical impedance spectroscopy feature vector exceeds the preset feature value threshold range;

[0010] S4. If the electrochemical impedance spectroscopy feature vector exceeds the preset feature value threshold range, the electrochemical impedance spectroscopy feature vector is processed by Fourier transform to obtain the harmonic component distribution in the frequency domain.

[0011] S5, the harmonic component distribution is divided into multiple categories to obtain a health status classification label, which includes a category that can be reused in stages and a category that needs to be dismantled and recycled.

[0012] S6, Generate a corresponding material property mapping table through the health status classification label, and obtain the remaining capacity estimate and cycle life estimate from the material property mapping table;

[0013] S7, determine whether the remaining capacity estimate and the cycle lifetime estimate are higher than a preset service threshold;

[0014] S8. If the estimated remaining capacity and the estimated cycle life are higher than the preset business threshold, then similar waste lithium batteries are grouped to obtain an optimized sorting batch sequence.

[0015] S9, send control commands to the automated conveying system according to the optimized sorting batch sequence.

[0016] Furthermore, step S1 includes:

[0017] Step S11: Place the waste lithium battery in a pulsed voltage field generated by an external signal generator, wherein the pulsed voltage field applies alternating stimulation to the positive and negative electrodes of the waste lithium battery;

[0018] Step S12: Monitor the current response and voltage fluctuation of waste lithium batteries under pulse voltage stimulation, and collect initial response current and voltage change data.

[0019] Step S13: Preprocess the initial response current and voltage change data, filter out noise components, and output them to subsequent feature extraction.

[0020] Furthermore, step S2 includes:

[0021] Step S21: Convert the initial response current and voltage change data into a time-series signal matrix;

[0022] Step S22: Perform convolution and pooling operations on the time-series signal matrix;

[0023] Step S23: Obtain the electrochemical impedance spectroscopy feature vector from the output layer of the convolutional neural network.

[0024] Furthermore, step S23 includes: the feature information obtained after the time-series signal matrix undergoes multi-layer convolution and pooling processing is input into the fully connected layer of the network, the fully connected layer integrates and reduces the dimensionality of the feature information, and finally outputs a feature vector of fixed length.

[0025] Furthermore, step S4 includes:

[0026] Step S41: Apply the Fast Fourier Transform algorithm to the electrochemical impedance spectroscopy eigenvectors;

[0027] Step S42: Extract the harmonic component distribution in the frequency domain, including the fundamental component and higher harmonic components;

[0028] Step S43: Normalize the distribution of the harmonic components and output it to the support vector machine classifier.

[0029] Furthermore, step S41 includes:

[0030] The input feature vector is padded with zeros or truncated as necessary to make its length a power of 2.

[0031] Furthermore, step S5 includes:

[0032] Step S51: Input the harmonic component distribution into the support vector machine classifier;

[0033] Step S52: Map the harmonic component distribution to a high-dimensional space using the kernel function of the support vector machine classifier;

[0034] Step S53: Delineate the decision boundary in the high-dimensional space to obtain the health status classification label;

[0035] Step S54: Based on the preset category threshold associated with the health status classification label, confirm the categories that can be used in stages and the categories that need to be dismantled and recycled.

[0036] Furthermore, step S6 includes:

[0037] Step S61: Retrieve the mapping relationship matching the health status classification label from the pre-stored database;

[0038] Step S62: Locate the remaining capacity estimate and cycle life estimate in the material property mapping table;

[0039] Step S63: Output the remaining capacity estimate and the cycle lifetime estimate to the service threshold judgment stage.

[0040] Furthermore, step S8 includes:

[0041] Step S81: Use the remaining capacity estimate and the cycle lifetime estimate as clustering feature vectors;

[0042] Step S82: Apply a clustering algorithm to determine cluster centers and similar groups;

[0043] Step S83: Assign similar waste lithium batteries to corresponding batches according to the cluster centers;

[0044] Step S84: Sort the corresponding batches according to the priority of tiered utilization to obtain the optimized sorting batch sequence;

[0045] Step S85: The optimized sorting batch sequence is transmitted to the control instruction generation module.

[0046] Furthermore, step S9 includes:

[0047] Step S91: Parse the optimized sorting batch sequence to generate control instructions;

[0048] Step S92: Send the control command to the automated conveying system through the communication interface;

[0049] Step S93: Receive execution feedback data returned by the automated conveying system and verify the completion status of the sorting operation.

[0050] The technical solution provided by this invention has the following beneficial effects:

[0051] This invention discloses an intelligent sorting and directional dismantling method for lithium battery recycling. Addressing the challenges of low classification accuracy, insufficient processing efficiency, and difficulties in optimizing the recycling and dismantling paths for secondary use in the lithium battery recycling industry, this invention achieves accurate classification and efficient processing through a series of technical means. The invention applies a specific frequency pulse voltage to waste lithium batteries to collect their electrochemical response data. Impedance spectrum features are extracted using a convolutional neural network, and a support vector machine classifier is used to classify the health status into multiple categories, generating classification labels. Then, based on the remaining capacity and cycle life estimates, the sorting batch sequence is optimized, and finally, control commands are sent to automated conveying equipment to achieve automated sorting. The core innovation of this invention lies in the deep integration of electrochemical analysis, machine learning, and automated control, significantly improving classification accuracy and processing efficiency, reducing the risk of misjudgment, optimizing resource allocation, and extending the overall lifespan of the secondary use system, providing an efficient and intelligent solution for the lithium battery recycling industry. Attached Figure Description

[0052] Figure 1 This is a flowchart of an intelligent sorting and directional dismantling method for lithium battery recycling according to the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] like Figure 1 As shown, this invention provides an intelligent sorting and targeted dismantling method for lithium battery recycling, aiming to achieve accurate classification and efficient processing of waste lithium batteries through a series of technical steps. The implementation process of this invention is described in detail below with reference to specific embodiments to make the objectives, technical solutions, and advantages of this invention clearer. In one embodiment, the method provided by this invention is mainly aimed at the recycling of waste lithium batteries. By analyzing the electrochemical characteristics of the batteries and combining intelligent algorithms and automated control technology, it achieves the assessment, sorting, and subsequent processing of battery health status. Waste lithium batteries can originate from electric vehicles, portable electronic devices, or energy storage systems, etc. These batteries experience performance degradation during use due to increased cycle counts or environmental factors, thus requiring scientific sorting and recycling strategies.

[0055] This embodiment will begin with signal acquisition and describe step by step how to classify and process waste lithium batteries using technical means. S1, Apply a pulse voltage stimulus of a specific frequency to the waste lithium battery to obtain the initial response current and voltage change data generated by the waste lithium battery.

[0056] In this step, the spent lithium batteries are placed on a specially designed testing platform equipped with a high-precision external signal generator. The signal generator produces pulsed voltage signals of specific frequency and amplitude, which are applied to the positive and negative electrodes of the battery to excite the electrochemical reactions within the battery. In this way, the dynamic response characteristics of the battery to external stimuli can be observed. The frequency range of the signal generator can be adjusted according to the battery type and testing requirements. For example, for lithium-ion batteries, a frequency range of 0.1 Hz to 1000 Hz is typically chosen to cover the electrochemical processes at different time scales within the battery. After the pulsed voltage is applied, the battery produces a corresponding current response, and the voltage value also fluctuates. These data are recorded in real time by a high-precision data acquisition device, forming initial response current and voltage change data. This data reflects the electrochemical impedance characteristics within the battery, laying the foundation for subsequent feature extraction and analysis. Specifically, during step S1, the testing platform needs to ensure a stable electrode connection between the spent lithium batteries and the signal generator to avoid data distortion due to poor contact. Data acquisition equipment typically includes high-sampling-rate current and voltage sensors, which can record the battery's response signals at millisecond intervals. Furthermore, to improve data reliability, the test environment needs to be controlled under constant temperature and humidity conditions, such as maintaining a temperature of 25 degrees Celsius and humidity at around 50%, to reduce interference from external environmental factors on the battery response. The acquired initial response current and voltage change data are usually stored in time-series format, with each set of data containing a timestamp, current value, and voltage value. This data will be used for further signal processing and feature extraction in subsequent steps.

[0057] Optionally, this step further includes: step S11, placing the waste lithium battery in a pulsed voltage field generated by an external signal generator, wherein the pulsed voltage field applies alternating stimulation to the positive and negative electrodes of the waste lithium battery.

[0058] In this sub-step, the spent lithium batteries are fixed to a specialized fixture on the testing platform, with the two ends of the fixture connected to the positive and negative output ports of a signal generator, respectively. The signal generator generates a pulsed voltage field that acts on the battery in an alternating manner. This alternating stimulation simulates the charging and discharging process of the battery in actual use, thereby triggering electrochemical reactions within the battery. The waveform of the alternating stimulation can be a square wave, sine wave, or triangular wave, depending on the testing objective. For example, square wave signals are suitable for fast response testing, while sine wave signals are more suitable for analyzing the battery's impedance characteristics. The amplitude of the pulse voltage is usually controlled within a safe range, such as between 0.5 volts and 2 volts, to avoid irreversible damage to the battery. The frequency and period of the alternating stimulation can also be adjusted according to the battery specifications; for example, for large-capacity power batteries, a lower frequency can be selected to capture their slow response characteristics. In one possible implementation, the testing platform can be equipped with a multi-channel signal generator to simultaneously apply pulsed voltage stimulation to multiple spent lithium batteries, thereby improving testing efficiency. Each battery's positive and negative terminals are connected to an independent signal channel, and the signal generator outputs different pulse parameters to each channel through program control. For example, for a batch of used lithium batteries from electric vehicles, different frequency combinations can be set, such as 0.1 Hz, 1 Hz, and 10 Hz, to test the battery's response characteristics at low, medium, and high frequencies, respectively. This setup can comprehensively reflect the battery's electrochemical behavior under different operating conditions, providing richer data support for subsequent classification. Step S12: Monitor the current response and voltage fluctuation of the used lithium batteries under pulse voltage stimulation, and collect initial response current and voltage change data.

[0059] In this sub-step, the testing platform monitors the battery's response signals in real time using high-precision current and voltage sensors. The current sensor is typically installed in the battery's positive terminal circuit to measure the instantaneous current change under pulsed voltage stimulation; the voltage sensor is connected in parallel between the battery's terminals to record voltage fluctuations. The sensor sampling frequency needs to be sufficiently high, for example, exceeding 1000 times per second, to capture subtle changes in the battery's response. The acquired data is transmitted to a computer system via a data acquisition card. The computer system performs preliminary storage and visualization processing on the data, allowing operators to observe the battery's response curve in real time. For example, during monitoring, operators can view the battery's current response curve and voltage fluctuation curve through the computer interface. If an anomaly is detected in the response curve of a battery, such as a sudden jump in current or a voltage value exceeding the normal range, the system automatically marks the battery as a potential problem sample and records its abnormal data. This abnormal data can be treated separately as special cases in subsequent analysis. During monitoring, the system also records environmental parameters, such as temperature and humidity during testing, to ensure consistent data acquisition conditions. The acquired initial response current and voltage change data are typically stored in a database in time series format. Each record contains three items: time point, current value, and voltage value, providing raw material for subsequent signal processing. Step S13 involves preprocessing the initial response current and voltage change data, filtering out noise components, and then outputting the data for subsequent feature extraction.

[0060] In this sub-step, the collected raw data often contains noise interference, such as electromagnetic interference from testing equipment or random fluctuations in the environment. This noise can affect the accuracy of subsequent analysis. Therefore, data preprocessing is necessary to improve its quality. The preprocessing process mainly includes two steps: data smoothing and noise filtering. Data smoothing can be achieved using the moving average method, which takes the average of several points before and after each point in the time series data to reduce short-term random fluctuations. Noise filtering can be achieved using digital filters; for example, a low-pass filter can effectively remove high-frequency noise while retaining the main trend information of the battery response. The preprocessed data is stored in a new time series format for use in subsequent feature extraction steps. In one embodiment, the preprocessing process may also include data normalization, which adjusts the numerical range of current and voltage data to a uniform interval, for example, mapping all data values ​​to between 0 and 1. Such processing helps reduce the impact of differences in data units between different batteries, improving the convergence speed and accuracy of subsequent algorithms. For example, for a batch of spent lithium batteries with varying capacities and specifications, the current values ​​in the raw data may fluctuate between milliamperes and amperes, and the voltage values ​​may vary between 2 volts and 4 volts. Normalization can unify these data into a standardized range, facilitating subsequent unified analysis. The preprocessed data is saved in file format and labeled with a unique identifier for each battery for traceability in subsequent steps. S2, Feature extraction is performed on the initial response current and voltage change data to obtain the electrochemical impedance spectroscopy feature vector.

[0061] In this step, the preprocessed time-series data is input into a pre-trained convolutional neural network (CNN) model specifically designed to extract feature information related to the battery's electrochemical characteristics from the time-series signal. Through multiple convolutional and pooling operations, the CNN automatically learns local patterns and global trends in the data, generating feature vectors that characterize the battery's impedance properties. These feature vectors reflect the battery's electrochemical response characteristics at different frequencies and can be used for subsequent health status assessments. Specifically, the input layer of the CNN receives preprocessed time-series data, organized into a matrix for easier network processing. The network's convolutional layers filter the data through a series of convolutional kernels, extracting time patterns related to battery impedance, such as periodic changes in current response or specific frequency components of voltage fluctuations. The pooling layers reduce the dimensionality of the convolutional layer outputs, decreasing the amount of data while retaining key feature information. After multiple convolutional and pooling operations, the network's output layer generates a fixed-length feature vector, namely the electrochemical impedance spectral feature vector. This feature vector contains impedance information of the battery at different frequencies, which can reflect the internal electrochemical state of the battery, such as the degree of degradation of electrode materials or electrolyte loss.

[0062] Optionally, this step further includes: step S21, converting the initial response current and voltage change data into a timing signal matrix.

[0063] In this sub-step, the preprocessed current and voltage data are organized into a two-dimensional matrix for easy input into the convolutional neural network. Each row of the matrix represents data at a specific time point, containing the current and voltage values ​​at that time; each column represents the complete change process of a time series. In this way, the raw data is converted into a structured form suitable for network processing. During the conversion process, the system checks the integrity of the data to ensure there are no missing or outlier values. If problematic data is found, it is automatically removed or interpolated. For example, when processing data from a batch of waste lithium batteries, the system may convert the test data of each battery into a 1000-row, 2-column matrix, where the 1000 rows correspond to 1000 time points, and the two columns correspond to the current and voltage values, respectively. Such a matrix structure can clearly reflect the dynamic response characteristics of the battery during the test process, making it easier for the network to extract feature information. The converted time-series signal matrix is ​​stored in a temporary file and labeled with the battery's identification number for matching and retrieval in subsequent steps. Step S22: Perform convolution and pooling operations on the time-series signal matrix.

[0064] In this sub-step, the time-series signal matrix is ​​input into the input layer of the convolutional neural network (CNN), which extracts features from the matrix using multiple convolutional kernels. Each convolutional kernel scans a local region within the matrix, extracting pattern information related to the battery's response characteristics, such as the periodicity of current changes or specific frequency components of voltage fluctuations. After convolution, the network performs dimensionality reduction on the feature map using pooling operations, such as max pooling or average pooling, to reduce data dimensionality while retaining key feature information. The combination of multiple convolutional and pooling operations progressively extracts feature information from low to high levels, ultimately forming a comprehensive characterization of the battery's electrochemical properties. In one possible implementation, the structure of the CNN can be adjusted according to testing requirements. For example, for scenarios with large amounts of data, the number of convolutional layers can be increased to improve feature extraction capabilities; for scenarios with limited computational resources, the stride of the pooling layers can be reduced to retain more detailed information. For example, when processing response data from spent lithium batteries in electric vehicles, a network structure containing three convolutional layers and two pooling layers can be designed. The first convolutional kernel mainly extracts short-term current fluctuation features, the second convolutional kernel extracts medium-term voltage change trends, and the third convolutional kernel integrates the feature information from the first two layers to form a description of the overall impedance characteristics of the battery. Such a network structure can effectively capture the response characteristics of the battery at different time scales, providing reliable feature information for subsequent analysis. Step S23: Obtain the electrochemical impedance spectrum feature vector from the output layer of the convolutional neural network.

[0065] In this sub-step, the feature information obtained after the time-series signal matrix undergoes multi-layer convolution and pooling processing is input into the fully connected layer of the network. The fully connected layer integrates and reduces the dimensionality of the feature information, ultimately outputting a fixed-length feature vector, namely the electrochemical impedance spectroscopy feature vector. This feature vector typically contains multiple dimensions, each corresponding to the impedance characteristics of the battery within a certain frequency range. For example, a feature vector of length 64 might contain impedance component information of the battery at low, medium, and high frequencies. This information reflects the internal electrochemical state of the battery, such as the degree of degradation of electrode materials or electrolyte loss. The output feature vector is saved in file format and labeled with the battery's identification number for use in subsequent steps. For example, when processing a batch of waste lithium batteries, the system might generate a feature vector of length 128 for each battery, where the first 32 dimensions correspond to low-frequency impedance characteristics, the middle 64 dimensions correspond to medium-frequency impedance characteristics, and the last 32 dimensions correspond to high-frequency impedance characteristics. By analyzing the distribution of these feature vectors, the health status of the battery can be preliminarily determined. For example, batteries with abnormal low-frequency impedance characteristics may have electrode material aging problems, while batteries with abnormal high-frequency impedance characteristics may have electrolyte loss problems. These feature vectors provide an important basis for subsequent threshold judgment and classification. S3, determine whether the electrochemical impedance spectroscopy feature vectors exceed the preset feature value threshold range.

[0066] In this step, the system compares the extracted electrochemical impedance spectroscopy feature vector with a preset feature value threshold range to determine whether the battery's impedance characteristics are within the normal range. The feature value threshold range is determined based on extensive historical data and experimental results, typically including the upper and lower limits of the feature values ​​for each dimension. For example, for a feature vector of length 128, a feature value threshold range can be set for each dimension, such as 0.2 to 0.8 for the first dimension, 0.1 to 0.9 for the second dimension, and so on. If the value of any dimension in the feature vector exceeds its corresponding feature value threshold range, the battery's impedance characteristics are considered abnormal, requiring further analysis of its frequency domain characteristics. Specifically, the feature value threshold judgment process is completed automatically by the computer system. The system checks each dimension value in the feature vector one by one and compares it with the preset feature value threshold range. If an abnormal dimension is found, the system records the specific value of that dimension and its corresponding feature value threshold range, and generates an anomaly report for operator reference. The anomaly report also includes the battery's identification number and the test time for subsequent traceability. Furthermore, the system can also preliminarily score the severity of battery anomalies based on the number and degree of anomalies. For example, batteries with more than 10 anomalies are rated as severely anomaly-prone, batteries with 5 to 10 anomalies are rated as moderately anomaly-prone, and batteries with fewer than 5 anomalies are rated as slightly anomaly-prone. These scores provide a reference for subsequent processing. In one embodiment, the feature value threshold range can be dynamically adjusted according to the battery type and application scenario. For example, for spent lithium batteries from electric vehicles, due to their complex working environment and high cycle count, a more lenient feature value threshold range can be set to avoid misjudgment; while for spent lithium batteries from portable electronic devices, due to their lower cycle count and relatively stable working environment, a stricter feature value threshold range can be set to improve detection accuracy. The process of dynamically adjusting the feature value threshold range can be achieved through historical data analysis. For example, the system will periodically statistically analyze the feature vector distribution of tested batteries over the past month and update the feature value threshold range based on the median and standard deviation of the distribution. This approach allows the feature value threshold range to better align with actual testing needs and improves the accuracy of judgment. S4. If the electrochemical impedance spectroscopy feature vector exceeds the preset feature value threshold range, the electrochemical impedance spectroscopy feature vector is processed by Fourier transform to obtain the harmonic component distribution in the frequency domain.

[0067] In this step, once the system detects an abnormal dimension in the electrochemical impedance spectroscopy (EIS) eigenvectors, it automatically triggers the frequency domain analysis process. Fourier transform is used here to convert the previously extracted time-domain eigenvectors to the frequency domain, allowing observation of the energy distribution of different frequency components in the battery response signal. The distribution of harmonic components in the frequency domain can more intuitively reflect the nonlinear characteristics of the internal electrochemical processes of the battery, such as electrode polarization, interfacial reaction hysteresis, or distortions caused by side reactions, which are often masked in the time-domain features. Specifically, the system first reassembles the EIS eigenvectors exceeding the eigenvalue threshold to meet the Fourier transform's input length requirements, and then applies the Fast Fourier Transform (FFT) algorithm to complete the conversion from the time domain to the frequency domain. After the conversion, the system extracts the main peak positions and corresponding amplitudes in the amplitude spectrum. These peak positions correspond to different harmonic frequencies, while the amplitude reflects the intensity of that frequency component. The harmonic component distribution obtained in this way includes the fundamental component and multiple harmonic components, revealing specific frequency response patterns that occur during battery health degradation. In one implementation, the harmonic component distribution exhibits significant differences for spent lithium batteries with varying degrees of degradation. For example, in normally or mildly degraded batteries, the harmonic components are mainly concentrated in the fundamental frequency and its lower harmonics, while the amplitudes of higher harmonics are typically very small. Severely degraded batteries, however, often show a significant enhancement of higher harmonics, especially a marked increase in the amplitudes of the 3rd and 5th harmonics. This phenomenon is usually associated with irreversible processes such as lithium dendrite growth, cathode structure collapse, or electrolyte decomposition. The system uses these frequency domain characteristics as an important basis for subsequent classification.

[0068] Optionally, this step also includes: step S41, applying a fast Fourier transform algorithm to the electrochemical impedance spectroscopy eigenvector.

[0069] In this sub-step, the system first performs necessary zero-padding or truncation on the input feature vector to ensure its length is a power of 2, thus meeting the data length requirements of the Fast Fourier Transform (FFT) algorithm. Next, the algorithm uses a divide-and-conquer strategy to continuously decompose the sequence into smaller subsequences, performing butterfly operations at each level to finally obtain a complex frequency domain representation. The system then takes the magnitude value of each frequency point to construct the amplitude spectrum, which serves as the main component of the harmonic component distribution. It should be noted that in actual testing, the original length of the feature vector is often determined by the output layer of the convolutional neural network, such as 128-dimensional or 256-dimensional. The system automatically selects the nearest power of 2 for zero-padding based on this length to minimize spectral leakage. After the FFT is completed, the system also removes the DC component, retaining only the frequency information corresponding to the AC component, thus focusing more on the harmonic characteristics in the battery's dynamic response. Step S42 extracts the harmonic component distribution in the frequency domain, including the fundamental component and higher harmonic components.

[0070] In this sub-step, the system locates the main peak positions in the amplitude spectrum and extracts the fundamental frequency (i.e., the dominant frequency of the stimulus pulse) and its harmonic components at integer multiples of that frequency in ascending order of frequency. Typically, the amplitude and phase information of the first 10 or 15 harmonics are extracted, forming a structured set of harmonic features. The fundamental frequency component primarily reflects the battery's linear response to the input stimulus, while higher harmonics characterize the severity of the nonlinear degradation process. For example, when processing a batch of spent lithium batteries from passenger vehicle power batteries, the system may find that the 5th harmonic amplitude of some batteries reaches more than 18% of the fundamental frequency amplitude, while that of others is less than 3%. The former often corresponds to batteries with more than 1200 cycles and a significant risk of internal short circuits, while the latter are mostly batteries with around 600 cycles and still exhibiting good consistency. This difference in harmonic amplitude proportions provides crucial distinguishing information for subsequent classification. Step S43: Normalize the distribution of the harmonic components and output it to the support vector machine classifier.

[0071] In this sub-step, the system performs maximum-min normalization or Z-score standardization on the extracted harmonic amplitudes to make the harmonic amplitudes of different batteries comparable. The normalized harmonic component distribution is organized into a fixed-dimensional vector, such as an 11-dimensional vector containing the fundamental, 2nd, 3rd...10th harmonic amplitudes, and then directly fed into the subsequent classification stage. Preferably, in some embodiments, phase information can also be further processed, such as calculating the phase difference of each harmonic relative to the fundamental wave and adding it as a supplementary feature to the vector to further enrich the input information of the classifier. The harmonic feature vector after normalization can effectively eliminate the dimensional influence caused by factors such as differences in battery capacity and small deviations in test voltage amplitude, improving classification consistency and robustness. S5, the harmonic component distribution is divided into multiple categories to obtain health status classification labels, which include categories that can be used in stages and categories that need to be dismantled and recycled.

[0072] In this step, the normalized harmonic component distribution vector is input into a pre-trained support vector machine (SVM) classification model. This model distinguishes batteries in different health states by constructing an optimal hyperplane in a high-dimensional space. The final output classification label clearly indicates whether the current battery is suitable for direct reuse or must undergo dismantling and recycling. Specifically, the SVM classifier uses a radial basis function kernel to map the original harmonic features to a high-dimensional feature space, where it searches for the classification hyperplane with the largest margin. A large number of labeled samples are used during the training phase, covering batteries from brand new to severely degraded throughout their entire lifespan. The labels are determined by professional testing institutions using a combination of capacity testing, internal resistance testing, and dissection analysis. In practical applications, the classifier can output two or more labels based on the input harmonic feature vector, with the two most crucial categories being "suitable for reuse" and "requires dismantling and recycling." In one possible implementation, the classifier can also output a confidence score, such as a continuous value between 0 and 1. A value closer to 1 indicates a stronger certainty that the battery belongs to the "reusable" category, while a value closer to 0 indicates a stronger certainty that it belongs to the "recyclable" category. The confidence score can provide a reference for subsequent manual review or automated decision-making. For example, when the confidence score is between 0.4 and 0.6, the system will automatically transfer the battery to the manual review queue to avoid resource waste caused by misjudgments in the model's boundary regions.

[0073] Optionally, this step also includes: step S51, inputting the harmonic component distribution into a support vector machine classifier.

[0074] In this sub-step, the normalized harmonic feature vector is directly fed into the classifier input layer via the data bus. The system first checks whether the vector dimension is consistent with that during training; if a mismatch occurs, an anomaly alarm is automatically triggered and logged. After the input check passes, the vector is sent to the kernel mapping module for high-dimensional space transformation. In step S52, the harmonic component distribution is mapped to a high-dimensional space using the kernel function of the support vector machine classifier.

[0075] In this sub-step, the radial basis function (RBF) performs a nonlinear mapping on the input vector according to preset parameters, transforming the originally linearly inseparable low-dimensional features into a linearly separable form in a high-dimensional space. The width parameter gamma of the kernel function is determined during the training phase through grid search and cross-validation, and typically ranges from 0.001 to 1, with the specific value depending on the distribution characteristics of the training samples. Step S53 involves delineating the decision boundary in the high-dimensional space to obtain the health status classification label.

[0076] In this sub-step, the Support Vector Machine (SVM) uses support vectors (i.e., the sample points closest to the hyperplane) to determine the position and orientation of the classification boundary. The decision function discriminates the input vectors and outputs category labels. For binary classification problems, the output is +1 or -1, corresponding to "reusable" and "recyclable" respectively. For extended multi-class scenarios, multiple binary classifiers are combined using a "one-to-many" or "directed acyclic graph" strategy. For example, when processing a batch of retired electric logistics vehicle battery packs, the system might classify approximately 42% of the batteries as "reusable," as these batteries have relatively low 3rd and 5th harmonic amplitudes and insignificant overall nonlinear characteristics; while the remaining 58% of the batteries are classified as "recyclable" due to significantly enhanced higher harmonics. These two types of batteries have significantly different subsequent processing paths; the former can directly enter the reusable production line, while the latter needs to enter the dismantling station for refined material recycling. Step S54: Based on the health status classification label associated with a preset category threshold, the reusable category and the recyclable category are confirmed.

[0077] In this sub-step, the system calls a pre-configured category threshold mapping table based on the classification label, converting the abstract +1 / -1 label into a category name with business meaning. Simultaneously, the system checks whether the confidence score reaches a preset confidence threshold; for example, a score above 0.75 is required to formally confirm the classification result; otherwise, it is marked as pending review. This dual confirmation mechanism further reduces the risk of misclassification. S6, a corresponding material property mapping table is generated using the health status classification label, and the remaining capacity estimate and cycle life estimate are obtained from the material property mapping table.

[0078] In this step, the system retrieves the corresponding capacity decay model and lifetime prediction parameters from a pre-built material property mapping database based on the health status classification labels obtained in the previous step. This allows it to calculate the remaining usable capacity percentage and estimated remaining cycle life of the current battery. These estimates will become the core basis for subsequent business threshold judgments and batch sorting optimization. Specifically, the material property mapping table is a large, offline-built lookup table that records the statistical correspondence between different health status categories and typical capacity retention rates and remaining cycle life percentages. This table is trained using a large amount of real-world data; for example, after conducting full-lifecycle tests on thousands of batteries of the same model, the system statistically analyzes the capacity retention rate distribution range for each health category at different cycle lifespans, and uses the median or lower confidence interval as the mapping value. In one implementation, for batteries marked as "reusable," the system may read from the mapping table an estimated remaining capacity of 68% to 85% of the initial capacity, corresponding to an estimated remaining cycle life of 22% to 41% of the initial cycle life. For batteries marked as "requiring dismantling and recycling," the estimated remaining capacity is typically less than 55% of the initial capacity, and the estimated remaining cycle life is less than 12% of the initial life. These numerical ranges are derived from the statistical patterns of actual test data and can reflect real degradation conditions well.

[0079] Optionally, this step also includes: step S61, retrieving from a pre-stored database a mapping relationship that matches the health status classification label.

[0080] In this sub-step, the system queries the material property mapping database using the health status classification label as the primary key. The database employs an efficient key-value storage structure, capable of returning matching records within milliseconds. If a label mismatch occurs (e.g., a rare degradation pattern appears), the system will invoke the default conservative estimate and record the event in the log system for subsequent model updates. Step S62: Locate the remaining capacity estimate and cycle life estimate in the material property mapping table.

[0081] In this sub-step, the system reads the capacity retention rate range and remaining cycle count range for the corresponding category based on the retrieved mapping relationship records. Typically, the mapping table stores the mean, lower limit, and confidence interval simultaneously. The system can select to use one of these columns based on business strategies. For example, in risk-sensitive tiered utilization scenarios, the lower limit is preferred as a conservative estimate; in resource maximization and recovery scenarios, the mean is preferred as a reference. Step S63 outputs the remaining capacity estimate and the cycle lifetime estimate to the business threshold judgment stage.

[0082] In this sub-step, the system transmits the two estimated values ​​read in structured data form to the subsequent judgment module, along with auxiliary information such as the battery's unique identifier, classification label, and confidence score, to ensure data traceability throughout the entire process. S7, determine whether the estimated remaining capacity and the estimated cycle life are higher than a preset business threshold.

[0083] In this step, the system compares the estimated remaining capacity and cycle life obtained in the previous step with pre-configured service thresholds. Only when both indicators are simultaneously above their respective service thresholds is the battery allowed to enter the secondary utilization sorting process; if either indicator is below its corresponding service threshold, it is deemed unsuitable for secondary utilization and must enter the dismantling and recycling path. Specifically, the remaining capacity threshold is typically set between 65% and 75% of the initial capacity, and the cycle life threshold is typically set between 18% and 30% of the initial cycle life. The specific values ​​are determined by the recycling company's business strategy, target application scenario, and safety margin requirements. For example, for backup power secondary utilization scenarios, the service thresholds can be set relatively leniently; while for communication base station backup power scenarios with higher requirements for consistency and lifespan, the service thresholds need to be more stringent. In one embodiment, the system supports a dynamic threshold strategy, that is, automatically adjusting the service thresholds according to the target use of the current recycling batch. For example, when this batch of batteries is mainly used for power replenishment of low-speed electric vehicles, the remaining capacity threshold can be lowered to 62%, and the cycle life threshold can be lowered to 15%; when the target is backup power for communication base stations, the remaining capacity threshold can be raised to 78%, and the cycle life threshold can be raised to 32%. Dynamic threshold adjustment can significantly improve resource utilization efficiency while ensuring the safety of use in different scenarios. S8, if the estimated remaining capacity and the estimated cycle life are higher than the preset business threshold, then similar waste lithium batteries are grouped to obtain an optimized sorting batch sequence.

[0084] In this step, the system performs similarity clustering on all "reusable" batteries that pass the business threshold assessment. Based on multiple dimensions such as remaining capacity, estimated lifespan, internal resistance increment, and consistency deviation, batteries with similar performance are grouped together, forming several sorting batches with similar performance. These batches are then sorted from high to low according to their reusability value, generating the final sorting batch sequence. Specifically, the clustering analysis uses the feature vector of each battery as input. This vector typically includes multiple dimensions such as estimated remaining capacity, estimated cycle life, classification confidence, internal resistance at the test temperature, and voltage plateau offset. The system uses mature clustering algorithms to find natural groupings, ensuring that the battery performance within a group is as consistent as possible, and the differences between groups are as significant as possible. After clustering, each cluster is assigned a batch number, and the average performance index of that batch is calculated for subsequent sorting and pricing reference. In one possible implementation, the system can also fine-tune the clustering results based on the actual application scenario. For example, for batteries planned for use in the same energy storage system, priority is given to ensuring the consistency of voltage plateau and internal resistance within the batch; for batteries planned for use in tiered scenarios with different power requirements, more attention is paid to the balanced distribution of remaining capacity and lifespan. This grouping method can significantly reduce the current imbalance problem when used in parallel later, and extend the overall lifespan of the tiered utilization system.

[0085] Optionally, this step further includes: step S81, using the remaining capacity estimate and the cycle lifetime estimate as clustering feature vectors.

[0086] In this sub-step, the system first constructs a clustering feature vector for each battery, typically including at least two core indicators: remaining capacity percentage and estimated remaining cycle count. Auxiliary indicators such as internal resistance growth rate, capacity decay curve slope, and temperature sensitivity coefficient can be added as needed. These features, after standardization, form the input matrix for clustering. Step S82 applies a clustering algorithm to determine cluster centers and similar clusters.

[0087] In this sub-step, the system runs a clustering algorithm, iteratively updating the center position of each cluster and assigning each battery to the nearest cluster center. The algorithm converges within a preset maximum number of iterations, or terminates early when the cluster center movement is less than a certain threshold. After clustering, the system outputs statistical information such as the center vector of each cluster, the number of batteries in the cluster, and the performance variance within the cluster. Step S83: Assign similar used lithium batteries to the corresponding batches based on the cluster centers.

[0088] In this sub-step, each battery is permanently marked with its cluster number, which becomes the subsequent batch number. The system also records the distance of each battery to the cluster center as a reference indicator for consistency within the batch. Batteries farther away are still assigned to the same cluster, but will receive lower priority in subsequent sorting. Step S84: Sort the corresponding batches according to their priority, obtaining the optimized sorting batch sequence.

[0089] In this sub-step, the system calculates a batch value score based on the comprehensive performance indicators of each cluster center. The score is typically a weighted combination of multiple indicators such as remaining capacity, remaining cycle life, and intra-cluster consistency. The batch with the highest score is placed at the front of the sequence as the highest-value tiered utilization batch; the batch with the lowest score, but still judged by the business threshold, is placed at the end of the sequence, usually used for backup power scenarios with lower performance requirements. Step S85: The optimized sorted batch sequence is transmitted to the control command generation module.

[0090] In this sub-step, the optimized sorting batch sequence is transmitted to the downstream control system in the form of a structured file or database record, providing an execution basis for automated conveying and sorting. S9, control commands are sent to the automated conveying system according to the optimized sorting batch sequence.

[0091] In this step, the system transforms the optimized sorting batch sequence into specific conveyor path planning and sorting instructions, which are then sent to the central controller of the automated conveyor system via industrial Ethernet or fieldbus. Upon receiving the instructions, the controller sequentially activates the corresponding conveyor lines, robotic arms, sorting gates, and other equipment according to the batch order, accurately delivering each battery to the designated secondary utilization buffer area or dismantling and recycling inlet. Specifically, the control instructions typically include information such as batch number, unique battery identifier, target buffer area number, conveying priority, and estimated arrival time. The automated conveyor system adjusts the belt speed, robotic arm gripping sequence, and sorting gate opening and closing status in real time based on these instructions, ensuring that high-value batches are processed first and low-value batches are queued in an orderly manner. In one implementation, the system also supports a real-time feedback mechanism. After completing each batch conveying task, the automated conveyor system returns an execution status, including information such as the number of successfully sorted batteries, the number of abnormal batteries, and the actual conveying time. This feedback data is recorded in a central database for subsequent process optimization and equipment maintenance decisions.

[0092] Optionally, this step may also include: step S91, parsing the optimized sorting batch sequence to generate control instructions.

[0093] In this sub-step, the instruction generation module reads the batch sequence line by line, fills in specific parameters according to a preset instruction template, and forms a standardized control message. For example, each instruction may include fields such as the starting conveyor line number, the target buffer area number, the expected number of batteries, and a priority flag. In step S92, the control instruction is sent to the automated conveying system through the communication interface.

[0094] In this sub-step, the system sends instructions to the conveyor system controller via protocols such as Industrial Ethernet, PROFIBUS, ModbusTCP, or OPCUA. The communication process employs a response confirmation mechanism to ensure that instructions are not lost or duplicated. Step S93 involves receiving execution feedback data from the automated conveyor system to verify the completion status of the sorting operation.

[0095] In this sub-step, the system monitors feedback messages from the conveyor system in real time, parsing the information contained within, such as task completion flags, exception codes, and completion timestamps. If an anomaly is detected in a batch sorting, such as mechanical jamming, battery recognition failure, or a full target buffer, the system will automatically pause subsequent instructions for that batch and trigger audible and visual alarms and manual intervention procedures. The sorting task is only considered complete when all batches return a successful completion status.

[0096] If the technical solution of this application involves the processing of personal information, the relevant products have established a sound user authorization mechanism: before collecting, using, or sharing personal information, the obligation to inform is fulfilled in accordance with the law, and the individual's voluntary and explicit consent is obtained; if sensitive personal information is involved, the user's separate and explicit consent is further obtained. Specific measures include, but are not limited to: setting up prominent prompts in the information collection area, or clearly displaying the processing rules (including the processor, purpose, method, information type, etc.) through electronic interfaces such as pop-ups, checkboxes, and active submissions, to ensure that users voluntarily authorize based on their knowledge. All personal information processing activities strictly comply with national laws and regulations, especially the relevant provisions of the "Personal Information Protection Law of the People's Republic of China," to effectively safeguard the legitimate rights and interests of personal information subjects.

[0097] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the invention. Those skilled in the art will understand that implementing all or part of the above-described embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. An intelligent sorting and targeted disassembly method for lithium battery recycling, characterized in that, include: S1, apply a pulse voltage stimulus of a specific frequency to the waste lithium battery to obtain the initial response current and voltage change data generated by the waste lithium battery; S2, feature extraction is performed on the initial response current and voltage change data to obtain the electrochemical impedance spectroscopy feature vector; S3, determine whether the electrochemical impedance spectroscopy feature vector exceeds the preset feature value threshold range; S4. If the electrochemical impedance spectroscopy feature vector exceeds the preset feature value threshold range, the electrochemical impedance spectroscopy feature vector is processed by Fourier transform to obtain the harmonic component distribution in the frequency domain. S5, the harmonic component distribution is divided into multiple categories to obtain a health status classification label, which includes a category that can be reused in stages and a category that needs to be dismantled and recycled. S6, Generate a corresponding material property mapping table through the health status classification label, and obtain the remaining capacity estimate and cycle life estimate from the material property mapping table; S7, determine whether the remaining capacity estimate and the cycle lifetime estimate are higher than a preset service threshold; S8. If the estimated remaining capacity and the estimated cycle life are higher than the preset business threshold, then similar waste lithium batteries are grouped to obtain an optimized sorting batch sequence. S9, send control commands to the automated conveying system according to the optimized sorting batch sequence.

2. The method of claim 1, wherein, Step S1 includes: Step S11: Place the waste lithium battery in a pulsed voltage field generated by an external signal generator, wherein the pulsed voltage field applies alternating stimulation to the positive and negative electrodes of the waste lithium battery; Step S12: Monitor the current response and voltage fluctuation of waste lithium batteries under pulse voltage stimulation, and collect initial response current and voltage change data. Step S13: Preprocess the initial response current and voltage change data, filter out noise components, and output them to subsequent feature extraction.

3. The method of claim 1, wherein, Step S2 includes: Step S21: Convert the initial response current and voltage change data into a time-series signal matrix; Step S22: Perform convolution and pooling operations on the time-series signal matrix; Step S23: Obtain the electrochemical impedance spectroscopy feature vector from the output layer of the convolutional neural network.

4. The method of claim 3, wherein, Step S23 includes: the feature information obtained after the time-series signal matrix is ​​processed by multiple convolution and pooling is input into the fully connected layer of the network, the fully connected layer integrates and reduces the dimensionality of the feature information, and finally outputs a feature vector of fixed length.

5. The method of claim 1, wherein, Step S4 includes: Step S41: Apply the Fast Fourier Transform algorithm to the electrochemical impedance spectroscopy eigenvectors; Step S42: Extract the harmonic component distribution in the frequency domain, including the fundamental component and higher harmonic components; Step S43: Normalize the distribution of the harmonic components and output it to the support vector machine classifier.

6. The method of claim 5, wherein, Step S41 includes: The input feature vector is padded with zeros or truncated as necessary to make its length a power of 2.

7. The method of claim 1, wherein, Step S5 includes: Step S51: Input the harmonic component distribution into the support vector machine classifier; Step S52: Map the harmonic component distribution to a high-dimensional space using the kernel function of the support vector machine classifier; Step S53: Delineate the decision boundary in the high-dimensional space to obtain the health status classification label; Step S54: Based on the preset category threshold associated with the health status classification label, confirm the categories that can be used in stages and the categories that need to be dismantled and recycled.

8. The method as described in claim 1, characterized in that, Step S6 includes: Step S61: Retrieve the mapping relationship matching the health status classification label from the pre-stored database; Step S62: Locate the remaining capacity estimate and cycle life estimate in the material property mapping table; Step S63: Output the remaining capacity estimate and the cycle lifetime estimate to the service threshold judgment stage.

9. The method as described in claim 1, characterized in that, Step S8 includes: Step S81: Use the remaining capacity estimate and the cycle lifetime estimate as clustering feature vectors; Step S82: Apply a clustering algorithm to determine cluster centers and similar groups; Step S83: Assign similar waste lithium batteries to corresponding batches according to the cluster centers; Step S84: Sort the corresponding batches according to the priority of tiered utilization to obtain the optimized sorting batch sequence; Step S85: The optimized sorting batch sequence is transmitted to the control instruction generation module.

10. The method as described in claim 1, characterized in that, Step S9 includes: Step S91: Parse the optimized sorting batch sequence to generate control instructions; Step S92: Send the control command to the automated conveying system via the communication interface; Step S93: Receive execution feedback data returned by the automated conveying system to verify the completion status of the sorting operation.

Images