A lithium ion battery mechanical deformation estimation method based on mechanical deformation decoupling

Abstract

The application discloses a mechanical deformation decoupling-based lithium ion battery mechanical deformation estimation method. The method comprises the following steps: collecting lithium ion battery voltage data, and preprocessing the collected lithium ion battery voltage data; constructing a mechanical deformation decoupling mechanism; constructing a CNN non-recoverable mechanical deformation feature extraction framework; and constructing a Seq2Seq-GRU collaborative mechanical deformation estimation mechanism. The method can estimate mechanical deformation by using existing collected signals of a battery management system, reduces the dependence on external mechanical sensors, pressure sensors or displacement sensors, thereby reducing the system integration cost and engineering deployment difficulty; can respectively depict normal dynamic expansion / contraction in a charging and discharging cycle and long-term non-recoverable expansion in an aging process, thereby avoiding mutual interference of the two types of signals with different time scales; and can represent the current aging degree of the battery by using voltage platform offset, slope change and polarization enhancement, and improve the full life cycle estimation precision.

Description

Technical Field

[0001] This application relates to the field of battery mechanical deformation estimation technology, and in particular to a method for estimating the mechanical deformation of lithium-ion batteries based on mechanical deformation decoupling. Background Technology

[0002] Lithium-ion batteries undergo significant but often difficult-to-observe mechanical deformation during charging and discharging. Changes in thickness, expansion displacement, pressure-related deformation, or strain-related changes resulting from both the dynamic charging / discharging process and aging degradation all fall under the category of battery mechanical deformation. Battery mechanical deformation is directly related to battery life and safety risks. Periodic mechanical deformation during normal charging and discharging reflects the recoverable response of lithium insertion and extraction in electrode materials, while long-term accumulated irrecoverable mechanical deformation typically reflects aging phenomena such as internal side reactions, interfacial film thickening, damage to active material structures, and increased impedance. When irrecoverable mechanical deformation continues to increase or its growth rate rises abnormally, it may lead to abnormal module preload, uneven electrode pressure, casing bulging, electrolyte leakage, and even an increased risk of thermal runaway. Therefore, estimating battery mechanical deformation without significantly increasing hardware costs can provide crucial information for battery health management, mechanical safety margin assessment, and early safety warnings.

[0003] Existing technologies for monitoring battery mechanical deformation include direct sensor measurement, physical model-based estimation, and data-driven estimation. Direct sensor measurement typically involves arranging strain gauges, displacement sensors, pressure sensors, or fiber optic sensors within the battery casing, pouch surface, or fixture structure to acquire data on surface deformation, thickness changes, or stacking pressure variations. The advantage of this method is that the measurement results are intuitive and directly reflect the mechanical response at the measured location. Therefore, it is widely used in laboratory mechanism studies and some controlled testing platforms; however, its application in practical engineering is limited.

[0004] Physical model-based estimation methods typically calculate the expansion, pressure, or mechanical response of a battery during charging and discharging by constructing equivalent spring-damping models, viscoelastic relaxation models, stacked pressure models, or thermo-mechanical-electrical coupling models. These methods generally require inputs such as state of charge (SOC), temperature, rate capability, preload pressure, material elastic parameters, geometric dimensions, and boundary constraints, and the model parameters are obtained through experimental calibration. While physical models offer some mechanistic interpretation, their accuracy is highly dependent on the quality of parameter identification and consistency with operating conditions.

[0005] Data-driven approaches utilize machine learning or deep learning to establish mapping relationships between electrical signals, state variables, and mechanical responses. Existing methods include training regression models using multi-source signals such as voltage, current, temperature, SOC, SOH, pressure, or strain, while others employ models like DNN, LSTM, GRU, Transformer, Gaussian process regression, or support vector machines to estimate battery pressure, expansion, or related health states. These methods are more likely to fit complex nonlinear relationships compared to purely physical models, but they often rely on multi-source inputs, artificial features, or a large number of labeled samples.

[0006] However, existing methods typically model battery mechanical deformation as a single, holistic quantity, without adequately distinguishing between two fundamentally different sources of deformation: one is recoverable mechanical deformation caused by lithium insertion / extraction, which can be repeatedly recovered within a single charge-discharge cycle; the other is irrecoverable mechanical deformation caused by SEI film growth, loss of active materials, microcrack propagation, and accumulation of side reactions, which continuously increases with the number of cycles. Because these two types of mechanical deformation differ significantly in their driving mechanisms, rates of change, and time scales, directly modeling the total mechanical deformation in a unified manner can easily lead to interference between short-term charge-discharge fluctuations and long-term aging drift, thus affecting the accuracy of the full lifecycle estimation. Summary of the Invention

[0007] The main objective of this application is to provide a method for estimating the mechanical deformation of lithium-ion batteries based on mechanical deformation decoupling, so as to solve the technical problems in the prior art.

[0008] To achieve the above objectives, this application provides a method for estimating the mechanical deformation of lithium-ion batteries based on mechanical deformation decoupling, the method comprising: S10, Collect lithium-ion battery voltage data and preprocess the collected lithium-ion battery voltage data; S20, constructing a mechanical deformation decoupling mechanism; S30, Construct a CNN framework for extracting features of irreversible mechanical deformation; S40, construct a Seq2Seq-GRU collaborative mechanical deformation estimation mechanism.

[0009] Optionally, step S10 includes the following steps: S101, based on the voltage acquisition module in the battery testing system, battery testing cabinet or battery management system, acquires the terminal voltage sequence of lithium-ion battery during the cyclic charging and discharging process; S102, based on the BMS front-end sampling chip or voltage acquisition unit, obtains the voltage sequence of a single unit or module; S103, preprocess the terminal voltage sequence.

[0010] Optionally, the preprocessing includes outlier removal, smoothing and denoising, cyclic segmentation, interpolation resampling, unifying the number of sampling points, and normalization.

[0011] Optionally, step S20 includes: S201, the mechanical deformation value at the start of the first cycle of charging is used as a reference baseline; S202, for any subsequent cycle, the mechanical deformation offset relative to the reference baseline at the start of any said cycle is determined as an unrecoverable mechanical deformation; S203, determine the dynamic offset within any of the loops relative to the starting baseline of the loop as recoverable mechanical deformation, and smooth the starting baseline of each loop.

[0012] Optionally, step S30 includes: S301, input the preprocessed terminal voltage sequence into the CNN branch; S302, the CNN branch extracts local morphological features of the voltage curve through convolutional layers, pooling layers, flattening layers and fully connected layers to obtain the output value.

[0013] Optionally, step S40 includes: S401, the preprocessed terminal voltage sequence is input into the Seq2Seq-GRU encoder, the encoder reads each voltage sampling point in sequence and updates the hidden state, generating a context vector containing historical voltage dynamic information. S402 embeds the degradation prior feature vector output by the CNN in the third stage into the Seq2Seq-GRU network through feature concatenation, linear mapping, hidden state initialization or context fusion, so that the encoder or decoder carries the current aging stage information of the lithium-ion battery when processing the sequence. S403 is generated step by step by the Seq2Seq-GRU decoder based on the context vector and the degenerate prior, resulting in a sequence of recoverable mechanical deformations within the loop. S404, the recoverable mechanical deformation sequence is added to or fused with the unrecoverable mechanical deformation estimate output by the CNN branch to obtain the total mechanical deformation estimate of the lithium-ion battery.

[0014] Compared with the prior art, the beneficial effects of this application are as follows: The mechanical deformation estimation method for lithium-ion batteries based on mechanical deformation decoupling provided in this application uses the terminal voltage sequence as the core online input. This application can estimate mechanical deformation by utilizing the existing acquisition signals of the battery management system, reducing the dependence on external mechanical sensors, pressure sensors or displacement sensors, thereby reducing system integration costs and engineering deployment difficulties.

[0015] By decoupling the total mechanical deformation into recoverable and unrecoverable mechanical deformation, this application can separately characterize the normal dynamic expansion / contraction during the charge-discharge cycle and the long-term unrecoverable expansion during the aging process, thereby avoiding mutual interference between the two types of signals with different time scales.

[0016] By extracting features of irreversible mechanical deformation and degradation stages from the voltage curve morphology through CNN branches, this application can characterize the current aging degree of the battery using information such as voltage plateau shift, slope change and polarization enhancement, thereby improving the model's adaptability to nonlinear changes in the later stages of life and thus improving the accuracy of full life cycle estimation.

[0017] By embedding the degradation prior of the intermediate layer of a CNN into a Seq2Seq-GRU network, this application enables the sequence estimation model to perceive the current degradation stage when generating mechanical deformation curves, thereby improving the accuracy and consistency of voltage-mechanical deformation mapping under different aging stages.

[0018] By outputting recoverable mechanical deformation, unrecoverable mechanical deformation, and total mechanical deformation, this application can not only provide mechanical deformation estimation results, but also distinguish between normal charge and discharge deformation and aging cumulative deformation, providing a more interpretive basis for SOH auxiliary estimation and safety early warning.

[0019] Through collaborative technical processes, this application provides a low-cost, interpretable, and transferable mechanical deformation estimation scheme for power batteries, energy storage batteries, and laboratory battery management platforms, which helps improve battery mechanical safety monitoring and lifespan management capabilities. Attached Figure Description

[0020] Figure 1 A flowchart of a lithium-ion battery mechanical deformation estimation method based on mechanical deformation decoupling provided in this application embodiment.

[0021] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the application. Rather, these embodiments are provided to make the disclosure more thorough and complete, and to fully convey the scope of the disclosure to those skilled in the art.

[0023] To address the aforementioned technical problems, this application provides a method for estimating the mechanical deformation of lithium-ion batteries based on mechanical deformation decoupling. This method can be executed by a computer, such as... Figure 1 As shown, the method may include the following steps: S10, Collect lithium-ion battery voltage data and preprocess the collected lithium-ion battery voltage data.

[0024] In an exemplary embodiment, step S10 may include the following steps: S101, based on the voltage acquisition module in the battery testing system, battery testing cabinet or battery management system, acquires the terminal voltage sequence of lithium-ion battery during the cyclic charging and discharging process; S102, based on the front-end sampling chip or voltage acquisition unit of BMS (Battery Management System), obtains the voltage sequence of individual cells or modules; S103, preprocess the terminal voltage sequence.

[0025] Specifically, the preprocessing includes outlier removal, smoothing and denoising, cyclic segmentation, interpolation resampling, unifying the number of sampling points, and normalization.

[0026] Furthermore, the preprocessing treats each complete charge-discharge cycle as a sample and extracts a fixed number of voltage sampling points at equal intervals within each cycle. For cycle data of different lengths, linear interpolation, spline interpolation, or equal-capacity interval sampling can be used to map them to a uniform length. For noisy points, moving average, local smoothing, or threshold removal can be used for processing. For dimensional differences between different lithium-ion batteries or different operating conditions, min-max normalization or standardization can be used. Through the above preprocessing, different cycle samples are made comparable under a uniform data scale, and a consistent input is provided for subsequent mechanical deformation decoupling, CNN feature extraction, and Seq2Seq-GRU estimation.

[0027] The specific threshold rules for anomaly point removal are as follows: Sampling points that exceed the battery's safe operating boundary are directly removed: Individual cell voltages exceeding the range of [2.5V, 4.35V] and sampling temperatures exceeding the range of [-10℃, 60℃] are directly determined as invalid anomalies; Cyclic discharge capacities exceeding 0.7~1.1 times the rated capacity are directly marked as anomalies.

[0028] The outlier handling logic is as follows: single outliers are directly replaced with the median within the window; when the number of consecutive outliers does not exceed 3, linear interpolation of the effective values ​​on both sides is used to complete the sequence; if the number of consecutive outliers exceeds 5, the sequence is truncated to avoid invalid data interfering with subsequent feature extraction.

[0029] Furthermore, during the data preparation process for model building, early cycles with a relatively high state of harm (SOH) of the lithium-ion battery can be used as the training interval, while later cycles with a further decrease in SOH can be used as the test interval to verify the model's ability to extrapolate and estimate the state of harm before the aging phase is observed. It should be noted that this step pertains to data partitioning and model building preparation after data acquisition and preprocessing, and is not the online data acquisition step itself. In the online application phase, at least the input voltage sequence, mechanical deformation, capacity, current, temperature, and cycle count data are primarily used for offline modeling, label construction, or verifying model performance.

[0030] S20, construct a mechanical deformation decoupling mechanism.

[0031] It is important to understand that the total mechanical deformation of a battery is not a quantity from a single source, but rather the result of the superposition of recoverable and unrecoverable mechanical deformation. Among them, recoverable mechanical deformation is mainly caused by lithium insertion / deintercalation of the electrode active materials during the charge and discharge process, which manifests as a periodic increase or decrease within a single cycle with changes in voltage and state of charge; unrecoverable mechanical deformation is mainly caused by the growth of the SEI (solid electrolyte interface) film, damage to the active material structure, the formation of dead lithium, and the accumulation of side reactions, which manifests as a long-term baseline drift that gradually rises across cycles.

[0032] In an exemplary embodiment, step S20 may include the following steps: S201, the mechanical deformation value at the start of the first cycle of charging is used as a reference baseline; S202, for any subsequent cycle, the mechanical deformation offset relative to the reference baseline at the start of any said cycle is determined as an unrecoverable mechanical deformation; S203, the dynamic offset within any cycle relative to the starting baseline of that cycle is determined as recoverable mechanical deformation, and the starting baseline of each cycle is smoothed to reduce the influence of random noise on the unrecoverable mechanical deformation label. For example, a moving average of several adjacent cycle starting points is used as the unrecoverable mechanical deformation baseline, or a low-pass filter and trend fitting method is used to extract the long-term drift term. Through this mechanical deformation decoupling mechanism, the total mechanical deformation of the lithium-ion battery can be decomposed into two parts: a long-term aging baseline and short-term charge-discharge dynamics, allowing subsequent models to learn the change patterns at different time scales.

[0033] The reason for employing the aforementioned mechanical deformation decoupling mechanism is that if the total mechanical deformation of a lithium-ion battery is directly used to train the model, the network needs to simultaneously fit the rapid fluctuations within a single cycle and the long-term drift across cycles, which can easily lead to mutual interference between the two types of changes. After decoupling, irreversible mechanical deformation can serve as a physical representation of the degradation stage of a lithium-ion battery, while recoverable mechanical deformation is used to describe the dynamic response within the current cycle. Therefore, the model not only outputs the total mechanical deformation but can also explain how much of this total comes from the normal charge-discharge process and how much comes from the aging accumulation process.

[0034] S30, Construct a framework for extracting features of irreversible mechanical deformation using CNN (Convolutional Neural Networks).

[0035] In an exemplary embodiment, step S30 may include the following steps: S301, input the preprocessed terminal voltage sequence into the CNN branch; S302, the CNN branch extracts local morphological features of the voltage curve through convolutional layers, pooling layers, flattening layers and fully connected layers to obtain the output value.

[0036] It is important to note that "local morphology feature extraction" here refers to the implementation method, while "irreversible mechanical deformation feature extraction" is the technical objective. Specifically, lithium-ion battery aging causes voltage plateau shifts, slope changes, polarization enhancement, and local peak shape changes. These local or multi-scale morphology changes are related to degradation mechanisms such as SEI film growth, active lithium loss, and increased impedance, which in turn drive the accumulation of irreversible mechanical deformation. Therefore, CNN extracts local morphology features from the terminal voltage curve to form a feature representation that can characterize irreversible mechanical deformation and degradation stages.

[0037] Specifically, the CNN network structure parameters are as follows: I. Input Layer Parameters Adapting the BMS-sampled terminal voltage time series, the preprocessed sliding window voltage series is reshaped into an input format suitable for 1D-CNN: Input shape: [batch size, sampling window length, 1], where the sampling window length is recommended to be set to 60~120 sampling points, corresponding to a 1~2 minute charge and discharge segment at a conventional BMS sampling frequency of 1Hz, which preserves the voltage dynamic characteristics without introducing too much redundant data.

[0038] Input feature dimension: Only the terminal voltage single-channel data normalized to the [0,1] interval is retained, without the need to introduce redundant features such as current and temperature, which fits the low hardware dependency characteristics of the solution.

[0039] II. Core Convolutional Layer Parameters A two-layer progressive convolutional structure is used to extract fine-grained features and long-term degradation features of the voltage sequence layer by layer: First convolutional layer: Kernel size: 3×1 (adapts to 1D time series data and avoids the unnecessary computational overhead of 2D convolution).

[0040] Number of convolution kernels: 32.

[0041] Step size: 1, edge padding set to same to ensure that the length of the sequence after convolution is consistent with the input.

[0042] Post-component: Paired with a batch normalization layer, using ReLU as the activation function, it introduces non-linear fitting capability and accelerates model training convergence.

[0043] Second convolutional layer: Kernel size: 3×1.

[0044] Number of convolution kernels: 64.

[0045] Step size: 1, edge fill: same.

[0046] Post-component: Also paired with batch normalization layer + ReLU activation to further extract deep voltage degradation correlation features.

[0047] III. Pooling and Regularization Layer Parameters While ensuring that core features are not lost, the feature dimensionality is significantly reduced, while avoiding overfitting issues under small sample training: Pooling layer: 2×1 max pooling is used with a stride of 2, which halves the length of the feature sequence output by the second convolution layer and directly compresses the number of parameters by 50%.

[0048] Dropout layer: Set the dropout probability p=0.2, randomly deactivate 20% of neurons during the training phase, enhance the model's generalization ability, and adapt to the distribution of battery data at different aging stages.

[0049] IV. Output Layer Parameters: The extracted voltage degradation features are mapped to degradation prior feature vectors that are adapted to subsequent Seq2Seq-GRU inputs: Flattened layer: Flattens the two-dimensional features output by pooling into a one-dimensional feature vector.

[0050] Fully connected layer: The output dimension is set to 64-dimensional, resulting in a 64-dimensional degenerate prior feature vector, which can be directly used as the initial hidden state of the subsequent Seq2Seq-GRU decoder, achieving seamless connection between CNN features and temporal deformation estimation model.

[0051] In this embodiment, the output value of the CNN branch can be an estimate of irreversible mechanical deformation. This output value is first used as a baseline estimate of the mechanical deformation caused by long-term aging of the lithium-ion battery, and is subsequently combined with the recoverable mechanical deformation to synthesize the total mechanical deformation. Simultaneously, the output of the intermediate layer of the CNN is selected as a degradation prior feature vector, preferably the output of the first fully connected layer, and this degradation prior feature vector is input into the Seq2Seq-GRU (sequence-to-sequence model based on gated recurrent units) branch in the fourth stage to constrain the mechanical deformation sequence estimation within the loop. In other words, the CNN branch not only provides a direct estimate of the irreversible mechanical deformation, but also provides prior information representing the current aging stage to the subsequent sequence estimation network.

[0052] The degradation prior feature vector can be understood as a digital representation of the current aging stage of a lithium-ion battery. For new batteries, the voltage curve plateau, slope, and polarization features are relatively stable; for aged batteries, the voltage curve may shift to the left, the slope may increase, polarization may be enhanced, and local morphology may change. After the CNN branch extracts features from these morphological changes, it can characterize the level of irreversible mechanical deformation accumulation and the degree of lithium-ion battery degradation. The reason for choosing intermediate layers instead of just using the final output layer is that the final output layer is usually compressed into a single prediction value, which may lose some degradation characterization information; while the first fully connected layer has already completed the global aggregation of convolutional features and retains richer voltage curve degradation morphology information, making it more suitable as a prior constraint for subsequent sequence estimation networks.

[0053] S40, construct a Seq2Seq-GRU collaborative mechanical deformation estimation mechanism.

[0054] In an exemplary embodiment, step S40 may include the following steps: S401, the preprocessed terminal voltage sequence is input into the Seq2Seq-GRU encoder, the encoder reads each voltage sampling point in sequence and updates the hidden state, generating a context vector containing historical voltage dynamic information. S402 embeds the degradation prior feature vector output by the CNN in the third stage into the Seq2Seq-GRU network through feature concatenation, linear mapping, hidden state initialization or context fusion, so that the encoder or decoder carries the current aging stage information of the lithium-ion battery when processing the sequence. S403 is generated step by step by the Seq2Seq-GRU decoder based on the context vector and the degenerate prior, resulting in a sequence of recoverable mechanical deformations within the loop. S404, the recoverable mechanical deformation sequence is added to or fused with the unrecoverable mechanical deformation estimate output by the CNN branch to obtain the total mechanical deformation estimate of the lithium-ion battery.

[0055] The aforementioned embodiments do not simply use Seq2Seq-GRU to directly fit voltage to mechanical deformation. Instead, they embed the degradation prior feature vector output by the CNN into the Seq2Seq-GRU, enabling the sequence model to perceive the current degradation stage of the lithium-ion battery when estimating the mechanical deformation within a cycle. Specifically, the embedding method can be to copy the degradation prior to each time step and concatenate it with the voltage features at the corresponding time point; alternatively, the degradation prior can be linearly mapped and used as part of the encoder's initial hidden state, the decoder's initial state, or the context vector; or the decoding process can be continuously modulated through gating or attention mechanisms. Through this embedding mechanism, Seq2Seq-GRU can avoid confusing the voltage-mechanical deformation relationship of new and aged batteries into the same mapping.

[0056] The aforementioned embodiments achieve technical synergy through a closed-loop process: "lithium-ion battery terminal voltage acquisition—preprocessing and sample construction—mechanical deformation decoupling mechanism—CNN-based extraction of irreversible mechanical deformation and degradation priors—Seq2Seq-GRU collaborative sequence estimation—mechanical deformation and early warning output." After completing the above collaborative estimation, the system can output irreversible mechanical deformation, recoverable mechanical deformation, and total mechanical deformation. Irreversible mechanical deformation mainly reflects the degree of long-term aging and irreversible expansion of the lithium-ion battery; recoverable mechanical deformation mainly reflects the dynamic expansion / contraction within the current charge-discharge cycle; and total mechanical deformation reflects the current overall mechanical change level of the lithium-ion battery. Furthermore, based on the total mechanical deformation amplitude, the cumulative amount of irreversible mechanical deformation, the mechanical deformation growth rate, and the degree of abnormal abrupt changes, the system can output SOH auxiliary estimation results, mechanical safety margin, abnormal expansion alarms, or maintenance suggestions. Thus, the output of the CNN branch in the third stage participates in the synthesis of total mechanical deformation and also serves as a degradation constraint for the sequence estimation in the fourth stage.

[0057] The mechanical deformation estimation method for lithium-ion batteries based on mechanical deformation decoupling provided in the above embodiments of this application can estimate mechanical deformation by using the terminal voltage sequence as the core online input. This reduces the dependence on external mechanical sensors, pressure sensors or displacement sensors, thereby reducing system integration costs and engineering deployment difficulties.

[0058] By decoupling the total mechanical deformation into recoverable and unrecoverable mechanical deformation, this application can separately characterize the normal dynamic expansion / contraction during the charge-discharge cycle and the long-term unrecoverable expansion during the aging process, thereby avoiding mutual interference between the two types of signals with different time scales.

[0059] By extracting features of irreversible mechanical deformation and degradation stages from the voltage curve morphology through CNN branches, this application can characterize the current aging degree of the battery using information such as voltage plateau shift, slope change and polarization enhancement, thereby improving the model's adaptability to nonlinear changes in the later stages of life and thus improving the accuracy of full life cycle estimation.

[0060] By embedding the degradation prior of the intermediate layer of a CNN into a Seq2Seq-GRU network, this application enables the sequence estimation model to perceive the current degradation stage when generating mechanical deformation curves, thereby improving the accuracy and consistency of voltage-mechanical deformation mapping under different aging stages.

[0061] By outputting recoverable mechanical deformation, unrecoverable mechanical deformation, and total mechanical deformation, this application can not only provide mechanical deformation estimation results, but also distinguish between normal charge and discharge deformation and aging cumulative deformation, providing a more interpretive basis for SOH auxiliary estimation and safety early warning.

[0062] Through collaborative technical processes, this application provides a low-cost, interpretable, and transferable mechanical deformation estimation scheme for power batteries, energy storage batteries, and laboratory battery management platforms, which helps improve battery mechanical safety monitoring and lifespan management capabilities.

[0063] In the description of this application, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0064] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the technical scope disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

[0065] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

Claims

1. A method for estimating the mechanical deformation of a lithium-ion battery based on mechanical deformation decoupling, characterized in that, The method includes the following steps: S10, Collect lithium-ion battery voltage data and preprocess the collected lithium-ion battery voltage data; S20, constructing a mechanical deformation decoupling mechanism; S30, Construct a CNN framework for extracting features of irreversible mechanical deformation; S40, construct a Seq2Seq-GRU collaborative mechanical deformation estimation mechanism.

2. The method for estimating the mechanical deformation of a lithium-ion battery based on mechanical deformation decoupling according to claim 1, characterized in that, Step S10 includes the following steps: S101, based on the voltage acquisition module in the battery testing system, battery testing cabinet or battery management system, acquires the terminal voltage sequence of lithium-ion battery during the cyclic charging and discharging process; S102, based on the BMS front-end sampling chip or voltage acquisition unit, obtains the voltage sequence of a single unit or module; S103, preprocess the terminal voltage sequence.

3. The method according to claim 2, characterized in that, The preprocessing includes outlier removal, smoothing and denoising, cyclic segmentation, interpolation resampling, unifying the number of sampling points, and normalization.

4. The method according to claim 1, characterized in that, Step S20 includes: S201, the mechanical deformation value at the start of the first cycle of charging is used as a reference baseline; S202, for any subsequent cycle, the mechanical deformation offset relative to the reference baseline at the start of any said cycle is determined as an unrecoverable mechanical deformation; S203, determine the dynamic offset within any of the loops relative to the starting baseline of the loop as recoverable mechanical deformation, and smooth the starting baseline of each loop.

5. The method according to claim 1, characterized in that, Step S30 includes: S301, input the preprocessed terminal voltage sequence into the CNN branch; S302, the CNN branch extracts local morphological features of the voltage curve through convolutional layers, pooling layers, flattening layers and fully connected layers to obtain the output value.

6. The method according to claim 1, characterized in that, Step S40 includes: S401, the preprocessed terminal voltage sequence is input into the Seq2Seq-GRU encoder, the encoder reads each voltage sampling point in sequence and updates the hidden state, generating a context vector containing historical voltage dynamic information. S402 embeds the degradation prior feature vector output by the CNN in the third stage into the Seq2Seq-GRU network through feature concatenation, linear mapping, hidden state initialization or context fusion, so that the encoder or decoder carries the current aging stage information of the lithium-ion battery when processing the sequence. S403 is generated step by step by the Seq2Seq-GRU decoder based on the context vector and the degenerate prior, resulting in a sequence of recoverable mechanical deformations within the loop. S404, the recoverable mechanical deformation sequence is added to or fused with the unrecoverable mechanical deformation estimate output by the CNN branch to obtain the total mechanical deformation estimate of the lithium-ion battery.

Images