Quality monitoring of battery cells using Raman spectroscopy

Abstract

A method (100) and a system (400) for determining the quality of a battery cell are disclosed. The method (100) includes the steps of: receiving (102), by a processor (502) of a control unit (406), from a spectrometer (402), an input Raman spectrum of the battery cell representing intensity values ​​for a wavelength range; comparing (104), by the processor (502), at least one feature extracted from the input Raman spectra with at least one model feature of a quality determination module; and, based on the comparison, determining (106), by the processor (502), whether the integrity state of the battery cell is faulty or fault-free. The system (400) includes a spectrometer (402) and a control unit (406) communicatively connected to the spectrometer (402).

Description

[0001] The present disclosure relates to the field of vehicle batteries. In particular, the present disclosure relates to techniques for determining the quality of battery cells using spectroscopy.

[0002] As is known in the art, the integrity state or quality status of a battery cell, battery pack, or module to be installed or embedded in a vehicle on an assembly line depends on various influencing factors, which generally include, for example, defects within the battery, the chemical composition of the battery cells / packs, or the like. All of these factors individually impact the quality and integrity of the battery cell in varying forms and degrees.

[0003] Furthermore, an efficient, real-time battery quality monitoring system is lacking on the assembly line immediately before the assembly of a vehicle, such as a car, etc. Current methods for evaluating battery quality during or after production, such as taking a voltage reading, measuring internal resistance using a pulse or AC impedance method, Coulomb counting, taking a snapshot of the battery's chemistry using electrochemical impedance spectroscopy (EIS), or the like, are often time-consuming, subjective, and prone to human error, leading to the installation of defective batteries, which poses safety risks and reduces the overall performance of the battery cell and the vehicle.

[0004] In addition, defects inside the battery, such as delamination, swelling, voids or cracks, are not easily visible from the outside, but can significantly affect the performance, safety and service life of the battery cell.

[0005] Furthermore, thickness variations, separator integrity, contaminants, etc., can affect the battery cell's energy storage capacity, power output, and overall efficiency. Detecting the presence of foreign objects or contaminants within the battery cell / battery pack, such as metal particles or foreign materials, is important because they can lead to internal short circuits, reduced performance, or even total failure. Detecting separator defects, such as cracks, wrinkles, or misalignment, is also critical to ensure battery cell safety and prevent internal short circuits.

[0006] Patent document US10502793B2 discloses a system and method for determining the physical condition of a battery, such as the state of charge (SOC), the state of integrity (SOH), the construction quality, a defect, or a fault condition. The method includes injecting two or more acoustic signals with two or more amplitudes, each acoustic signal having two or more frequencies, into the battery and detecting vibrations generated in the battery based on the two or more acoustic signals. Furthermore, nonlinear response characteristics of the battery for the two or more acoustic signals are determined from the detected vibrations. Subsequently, the physical condition of the battery is determined based at least in part on the nonlinear response characteristics using nonlinear acoustic resonance spectroscopy (NARS) or nonlinear resonant ultrasonic spectroscopy (NRUS).

[0007] While the conventional and cited prior art may disclose various systems and methods related to battery data analysis or monitoring methods, there is still room and need to provide an improved solution for determining the quality of a battery cell.

[0008] An object of the present disclosure is to provide a method and a system that overcomes the above-mentioned limitations of the existing systems and / or methods for determining the quality of battery cells using spectroscopy.

[0009] An object of the present disclosure is to provide techniques for efficient and real-time quality monitoring of battery cells in the assembly line.

[0010] Yet another object of the present disclosure is to provide techniques for determining an integrity state of the battery cell, which is a faulty or a non-faulty battery cell, using Raman spectroscopy.

[0011] Aspects of the present disclosure relate to the field of vehicle batteries. In particular, the present disclosure provides a method and system for determining the quality of battery cells using spectroscopy.

[0012] One aspect of the present disclosure relates to a method for determining the quality of a battery cell. The method may comprise the steps of: receiving, by a processor of a control unit from a spectrometer, an input Raman spectrum of the battery cell representing intensity values ​​for a wavelength range; comparing, by the processor, at least one feature extracted from the input Raman spectra with at least one model feature of a quality determination module; and, based on the comparison, determining, by the processor, whether the integrity state of the battery cell is faulty or fault-free.

[0013] Thus, the process supports efficient and accurate monitoring of battery cells for quality determination and fault detection in order to avoid problems during battery cell use.

[0014] In one aspect, the at least one feature may comprise intensity values ​​at one or more wavelengths, and the at least one model feature comprises one or more identified wavelengths and weights assigned to the one or more identified wavelengths.

[0015] In one aspect, the battery cell may be determined to be the faulty cell if the intensity values ​​at the one or more identified wavelengths are greater than a threshold.

[0016] According to a further aspect, the method for determining the quality of the battery cell can be carried out before the battery cell is installed in a vehicle.

[0017] According to one aspect, the quality determination module may be obtained by the steps of: collecting, by the processor using the spectrometer, a Raman spectrum data item from a plurality of sample battery cells, wherein the plurality of sample battery cells includes at least one of defective battery cells, non-defective battery cells, and at least one used battery cell; performing, by the processor, a baseline correction, a smoothing treatment, and a normalization treatment on the Raman spectrum data item to obtain preprocessed spectrum data; and obtaining, by the processor, the quality determination module by adopting the preprocessed spectrum data.

[0018] According to another aspect, the step of obtaining the quality determination module may also include the steps of: associating all Raman spectra of preprocessed spectrum data with the integrity state of the corresponding Raman spectra; splitting the preprocessed spectrum data into a training spectrum data item and a test spectrum data item; and inputting the training spectrum data item and the test spectrum data item into a machine learning model to obtain the quality determination module.

[0019] According to one aspect, the step of obtaining the quality determination module further comprises the step of: training the machine learning model, wherein the training includes: analyzing the training spectrum data item to identify the at least one model feature in the training spectrum data item.

[0020] In one aspect, the machine learning model may be a convolutional neural network (CNN).

[0021] The method therefore provides a root cause analysis of battery cell failures by using the Raman spectroscopy spectrum to monitor battery cells.

[0022] Another aspect of the present disclosure relates to a system for determining the quality of a battery cell. The system may include a spectrometer and a control unit communicatively coupled to the spectrometer. The control unit may further include a processor coupled to a memory capable of storing one or more instructions executable by the processor to: receive an input Raman spectrum of the battery cell representing intensity values ​​for a wavelength range from a spectrometer; compare at least one feature extracted from the input Raman spectra with at least one model feature of a quality determination module; and, based on the comparison, determine whether the integrity state of the battery cell is faulty or fault-free.

[0023] According to one aspect, to obtain the quality determination module, the control unit may be configured to: collect a Raman spectrum data item from a plurality of sample battery cells using the spectrometer, wherein the plurality of sample battery cells includes at least one of defective battery cells, non-defective battery cells, and at least one used battery cell; perform a baseline correction, a smoothing treatment, and a normalization treatment on the Raman spectrum data item to obtain preprocessed spectrum data; and obtain the quality determination module by adopting the preprocessed spectrum data.

[0024] According to one aspect, for the step of obtaining the quality determination module, the control unit may be further configured to: associate all Raman spectra of preprocessed spectrum data with the integrity state of the corresponding Raman spectra; divide the preprocessed spectrum data into a training spectrum data item and a test spectrum data item; and input the training spectrum data item and the test spectrum data item into a machine learning model to obtain the quality determination module.

[0025] According to a further aspect, the at least one feature may comprise intensity values ​​at one or more wavelengths, and the at least one model feature comprises one or more identified wavelengths and weights assigned to the one or more identified wavelengths.

[0026] In one aspect, the battery cell may be determined to be the faulty cell if the intensity values ​​at the one or more identified wavelengths are greater than a threshold.

[0027] According to one aspect, the control unit may be configured to determine the quality of the battery cell prior to installation of the battery cell in a vehicle, wherein the machine learning model may be a convolutional neural network (CNN).

[0028] The method and system for determining the quality of a battery cell enable early detection of defects in the battery cell and also help in creating a digital library based on the Raman spectrum for the battery cell.

[0029] Various objects, features, aspects and advantages of the inventive subject matter will become more apparent from the following detailed description of preferred embodiments together with the accompanying drawing figures, in which like reference numerals designate like components.

[0030] The accompanying drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification. The drawings illustrate exemplary embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure. Fig. 1 illustrates a flowchart depicting the proposed method for determining the quality of a battery cell in a vehicle assembly line according to an embodiment of the present disclosure. Fig. 2 illustrates an exemplary flowchart illustrating the operation of the proposed method for determining the quality of a battery cell in an assembly line according to an embodiment of the present disclosure. Fig. 3 illustrates another exemplary flowchart depicting the step-by-step operation of the proposed method for determining the quality of a battery cell in an assembly line according to an embodiment of the present disclosure. Fig. 4 illustrates an exemplary network architecture of the proposed system for determining the quality of a battery cell in a vehicle assembly line to illustrate its general operation according to an embodiment of the present disclosure. Fig. 5 illustrates exemplary functional units of a control unit in connection with the proposed system according to an embodiment of the present disclosure. Fig. 6A and Fig. 6B illustrate the operation of an image recognition module of the proposed system using a convolutional neural network (CNN) technique according to an embodiment of the present disclosure. Fig. 7A and Fig. 7B illustrate an exemplary spectrum of a faulty battery cell and an intact battery cell, each as detected by the proposed system, according to an embodiment of the present disclosure.

[0031] The following is a detailed description of the embodiments of the disclosure illustrated in the accompanying drawings. The embodiments are detailed enough to clearly convey the disclosure. However, the amount of detail provided is not intended to limit the expected variations of the embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosures as defined by the appended claims.

[0032] Embodiments discussed herein relate to the field of vehicle batteries. In particular, the present disclosure provides a method and system for determining the quality of battery cells using spectroscopy.

[0033] In one embodiment of the present disclosure, an artificial intelligence model and / or a machine learning model may be integrated to be trained and tested to determine the quality of a battery cell in an assembly line, wherein the model may include the steps to be performed in the following manner: Data collection: Collect a dataset of Raman spectra from a variety of sample battery cells, i.e., vehicle batteries, including defective, non-defective, and used examples. The Raman spectra represent the chemical composition and molecular structure of the battery cells.

[0034] Data preprocessing: Preprocessing the Raman spectra by removing noise, normalizing the data, and potentially applying feature extraction techniques to enhance the relevant information in the spectra.

[0035] Labeling the data: Annotating the Raman spectra with labels that indicate whether individual battery cells are intact or faulty. This step establishes the ground truth for training and validation.

[0036] Split into train-test: Split the labeled dataset into training and test sets. The training set is used to train a CNN model or module, while the test set is used to evaluate its performance.

[0037] Designing the CNN architecture: Designing the architecture of the CNN model, which typically consists of multiple convolutional layers for feature extraction, followed by fully connected layers for classification.

[0038] Training the model: Train the CNN model using the training set of Raman spectra (input). During training, the CNN model learns to identify patterns and features in the spectra that indicate healthy / defective battery cells (output).

[0039] Model validation: Evaluate the trained model using the test set. Measure performance metrics such as accuracy, precision, sensitivity, and F1 score to assess how well the CNN model generalizes to unseen data from the battery cell to be installed in the vehicle.

[0040] Deploying the model: Once the model meets the desired performance criteria, it is deployed for real-world application, so it should be able to accept new, unseen Raman spectra as input and classify them as good or faulty battery cells.

[0041] Real-time testing: In a production environment, Raman spectra of the battery cell are acquired in real time, and the acquired spectra are fed as input to the trained CNN model.

[0042] Defect detection: The CNN model then processes the Raman spectra and generates predictions for each battery cell, classifying them as intact or defective. The output can be in the form of a binary classification or a probability value indicating the likelihood of failure.

[0043] Validating the model: The reliability of the CNN model is ensured by continuously monitoring and validating the model's performance in a real-world environment and collecting feedback from the assembly line, comparing the model's predictions with the ground truth, and evaluating performance metrics such as accuracy, precision, and sensitivity.

[0044] Now with reference to Fig. 1, the proposed method 100 (also referred to herein as method 100) can be used to determine the quality of a battery cell in an assembly line for a vehicle.

[0045] The method 100 may comprise the steps of: at step 102, receiving, by a processor 502 of a control unit 406 from a spectrometer 402, an input Raman spectrum of the battery cell representing intensity values ​​for a wavelength range; at step 104, comparing, by the processor 502, at least one feature extracted from the input Raman spectra with at least one model feature of a quality determination module; and at step 106, based on the comparison, determining, by the processor (502), whether the integrity state of the battery cell is faulty or fault-free.

[0046] Further, the at least one feature may comprise intensity values ​​at one or more wavelengths, and the at least one model feature comprises one or more identified wavelengths and weights assigned to the one or more identified wavelengths.

[0047] The battery cell may be determined to be the faulty cell if the intensity values ​​at the one or more identified wavelengths are greater than a threshold value.

[0048] In one embodiment, these identified wavelengths correspond to a set of wavelengths whose weights are higher compared to other wavelengths, and these identified wavelengths can be used to determine the integrity state of the battery cell.

[0049] Furthermore, the method 100 for determining the quality of the battery cell can be performed before the battery cell is installed in a vehicle.

[0050] In one embodiment, the quality determination module may be obtained by: collecting a Raman spectrum data item from a plurality of sample battery cells; performing a baseline correction, a smoothing treatment, and a normalization treatment on the Raman spectrum data item to obtain preprocessed spectrum data; and obtaining the quality determination module by adopting the preprocessed spectrum data.

[0051] The plurality of example battery cells may include at least one of defective battery cells, non-defective battery cells, and used battery cells.

[0052] The quality determination module can be further obtained by: mapping all Raman spectra of preprocessed spectrum data to the integrity state of the corresponding Raman spectra; dividing the preprocessed spectrum data into a training spectrum data item and a test spectrum data item; and inputting the training spectrum data item and the test spectrum data item into a machine learning model to obtain the quality determination module.

[0053] The quality determination module can then be obtained by training the machine learning model. Here, training involves analyzing the training spectrum data item to identify at least one model feature in the training spectrum data item.

[0054] Furthermore, the machine learning model may be a convolutional neural network (CNN) model, which may be used to analyze the training spectrum data item. Furthermore, the machine learning model may also be any deep learning model, including, but not limited to, an artificial neural network model, a recurrent neural network model, a generative adversarial network model, a feedforward neural network model, or the like.

[0055] In a non-limiting example of the present disclosure, the step of training the machine learning model may be performed as follows: Once the input spectra (input Raman spectra) of thousands of battery cells (both good and faulty) have been generated using the spectrograph / spectrometer, they are used to train the model.

[0056] After preprocessing the spectral data and randomly splitting the training and test sets in an 80:20 ratio, this data is provided as input to the convolutional neural network to build a model for identifying faulty batteries. The training phase involves feature selection and dimensionality reduction. A given input feature can be highly relevant (contains information not present in any other feature), relevant, weakly relevant (contains some information present in other features), or irrelevant.

[0057] Feature selection involves selecting a subset of highly relevant features for use in model building. In this step, the one or more peaks and / or valleys are correlated with the measured properties / features. For example, peaks at 580 cm-1, 1240 cm-1, and 1460 cm-1 can be helpful in characterizing the errors in the battery. To expand the model's generalization capability, improve model robustness, and avoid overfitting, the model is evaluated using five-fold cross-validation. Cross-validation can be used to evaluate the model's real-world prediction error and tune the model's parameters.

[0058] For the machine learning model, the model further applies this learned knowledge of where to look for important, somewhat important, and unimportant features to classify new inputs based on the weights assigned to each feature. For example, since the model assigns a high weight to the peaks at 580 cm-1, 1240 cm-1, and 1460 cm-1 to indicate a faulty battery, any input with high intensity values ​​at these wavelengths will be classified as a faulty battery.

[0059] With reference to Fig. 2 is an exemplary flowchart illustrating the operation of the proposed method for determining the quality of a battery cell in an assembly line according to an embodiment of the present disclosure.

[0060] At block 202, a light beam may be focused onto the battery cell on the vehicle assembly line to receive the input spectra comprising the real-time spectrometric measurement data of the battery cell. These input spectra may be recorded as a series of intensity values ​​corresponding to different wavelengths, where the intensity values ​​correspond to the scattering of light as the light beam interacts with the battery cell. The light beam may be a laser light beam or the like.

[0061] Subsequently, continuing with block 204, the spectrometric measurements received in block 202 may be received as a Raman spectroscopy spectrum and then fed into the method 100 along with the one or more filtered example spectra of the plurality of example battery cells.

[0062] In block 206, the one or more data processing techniques may be used to determine the quality of the battery cell as output, where an input layer 206-1 may include the Raman spectroscopy spectrum and the one or more filtered example spectra, a hidden layer 206-2 may include the analysis of the Raman spectroscopy spectrum and the one or more filtered example spectra using the one or more data processing techniques.

[0063] Here, the one or more data techniques may include, but are not limited to, an AI model, an ML module, noise reduction and filtering, data normalization, feature extraction techniques for spectrum expansion, or the like.

[0064] At an output layer 206-3, the method 100 may then predict the quality of the battery cell as a good or faulty battery cell as an output. Here, the one or more data techniques may include a convolutional neural network (CNN) technique for processing the Raman spectroscopy spectrum, the one or more filtered example spectra, and / or the input spectra.

[0065] With reference to Fig. 3 is another exemplary flowchart 300 illustrating the step-by-step operation according to the proposed method for determining the quality of a battery cell in an assembly line, according to an embodiment of the present disclosure.

[0066] In block 302, a plurality of example battery cells may be supplied to the assembly line for a vehicle to perform the method 100 for determining the quality of the battery cell via a Raman spectroscopy technique in block 304.

[0067] Subsequently, in block 306, one or more data processing techniques, such as an AI model or a trained CNN model / module technique, may be utilized to perform an analysis of the input spectra of the plurality of example battery cells. The AI ​​model or the trained CNN model may be developed in block 306 to detect poor-quality or faulty battery cells.

[0068] In block 308, the plurality of sample battery cells may be tested, and then in block 310, the AI ​​model or trained CNN model may be developed using the test data from the plurality of sample battery cells. This trained CNN model or AI model may then be used to generate the battery cell quality status data on the vehicle assembly line. The trained model may also include a relationship between the input spectra and the battery cell quality.

[0069] Furthermore, the plurality of sample battery cells includes good-quality battery cells, used battery cells, and defective battery cells. The defects in the plurality of sample battery cells are generally artificially induced defects for training the CNN module to accurately generate the battery cell quality status.

[0070] With reference to Fig. Figure 4 illustrates an exemplary network architecture of the proposed system 400 (referred to interchangeably herein as system 400), wherein the proposed system 400 is used to determine the quality of a battery cell in a vehicle assembly line. The system 400 includes a spectrometer 402 and a controller 406 communicatively coupled to the spectrometer 402.

[0071] The control unit 406 may further include a processor 502 coupled to a memory 504 that may store one or more instructions executable by the processor 502 to: receive an input Raman spectrum of the battery cell representing intensity values ​​for a wavelength range from a spectrometer 402; compare at least one feature extracted from the input Raman spectra with at least one model feature of a quality determination module, and based on the comparison, determine whether the integrity state of the battery cell is faulty or fault-free.

[0072] To obtain the quality determination module, the control unit 406 may be further configured to: collect a Raman spectrum data item from a plurality of sample battery cells using the spectrometer 402; perform a baseline correction, a smoothing treatment, and a normalization treatment on the Raman spectrum data item to obtain preprocessed spectrum data; and obtain the quality determination module by adopting the preprocessed spectrum data.

[0073] The plurality of example battery cells may include at least one of defective battery cells, non-defective battery cells, and used battery cells.

[0074] The control unit 406 may be further configured to: associate each Raman spectrum of preprocessed spectrum data with the integrity state of the corresponding Raman spectra; divide the preprocessed spectrum data into a training spectrum data item and a test spectrum data item; and input the training spectrum data item and the test spectrum data item into a machine learning model to obtain the quality determination module.

[0075] The at least one feature may comprise intensity values ​​at one or more wavelengths, and the at least one model feature comprises one or more identified wavelengths and weights assigned to the one or more identified wavelengths.

[0076] In addition, the battery cell may be determined to be the faulty cell if the intensity values ​​at the one or more identified wavelengths are greater than a threshold value.

[0077] In one embodiment, these identified wavelengths may correspond to a set of wavelengths whose weights are higher compared to other wavelengths, and these identified wavelengths are used to determine the integrity state of the battery cell.

[0078] In another embodiment, the controller 406 may be configured to determine the quality of the battery cell prior to installing the battery cell in a vehicle, and the machine learning model may be a convolutional neural network (CNN).

[0079] In one embodiment, the system 400 also includes an actuator 404 coupled between the spectrometer 402 and the control unit 406. In an exemplary embodiment, a first signal sent from the control unit 406 may be received by the actuator 404, wherein the actuator 404, based on the received first signal, enables the activation of the spectrometer 402 and / or the system 400 to receive and analyze the input spectra comprising real-time spectrometric measurement data of the battery cell and to generate the quality status of the battery cell based on the analyzed input spectra. Furthermore, the actuator 404 may also enable the deactivation of the spectrometer 402 when not in use.

[0080] In an exemplary embodiment, the controller 406 is further configured to trigger a second signal, and upon receipt of the second signal, the actuator 404 assists in deactivating the spectrometer 402 or the system 400. The actuator 404 may enable selective activation and deactivation of the spectrometer 402 and / or the system 400 based on the detected first signal by the controller 406.

[0081] In one embodiment, the control unit 406 is communicatively coupled to the spectrometer 402 via a network 408. Further, the network 408 may be a wireless network, a wired network, or a combination thereof, which may be implemented as one of various network types, such as an intranet, local area network (LAN), wide area network (WAN), Internet, and the like. Furthermore, the network 408 may be either a dedicated network or a shared network. The shared network may represent an interconnection of various types of networks that may utilize a variety of protocols, for example, Hypertext Transfer Protocol (HTTP), Transmission Control Protocol / Internet Protocol (TCP / IP), Wireless Application Protocol (WAP), and the like.

[0082] In one embodiment, system 400 may be implemented using any one or a combination of hardware components and software components, such as a cloud, a server 410, a computing system, a computing device, a network device, and the like. Further, controller 406 interacts with spectrometer 402 via a website or application that may be located within proposed system 400. In one implementation, proposed system 400 may be accessed via a website or application that may be configured with any operating system, including, but not limited to, Android™, iOS™, and the like.

[0083] With reference to Fig. 5, block diagram 500 illustrates exemplary functional units of control unit 406 according to an embodiment of the present disclosure. Control unit 406 includes one or more processors 502. The one or more processors 502 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuitry, and / or any device that processes data based on operating instructions.

[0084] Among other capabilities, the one or more processors 502 are configured to retrieve and execute computer-readable instructions stored in a memory 504 of the controller 406. The memory 504 may store one or more computer-readable instructions or routines that may be retrieved and executed to create or share the data units over a network service. The memory 504 may include any non-transitory storage device, including, for example, volatile memory such as RAM, or non-volatile memory such as EPROM, flash memory, and the like.

[0085] In one aspect, the control unit 406 may also include one or more interfaces 506. The interface(s) 506 may include a variety of interfaces, for example, interfaces for data input and output devices, so-called I / O devices, storage devices, and the like.

[0086] The interface(s) 506 may support communication of the spectrometer 402 with various devices coupled to the control unit 406. The interface(s) 506 may also provide a communication path for one or more components of the control unit 406. Examples of such components include, but are not limited to, processing engine(s) 508 and a database 510.

[0087] In one aspect, the processing engine(s) 508 may be implemented as a combination of hardware and programming (e.g., programmable instructions) to implement one or more functionalities of the processing engine(s) 508. In the examples described herein, such combinations of hardware and programming may be implemented in various ways.

[0088] For example, the programming for the processing engine(s) 508 may consist of processor-executable instructions stored on a non-transitory machine-readable storage medium, and the hardware for the processing engine(s) 508 may include a processing resource (e.g., one or more processors) for executing such instructions.

[0089] In the present examples, the machine-readable storage medium may store instructions that, when executed by the processing resource, implement the processing engine(s) 508. In such examples, the controller 406 may include the machine-readable storage medium on which the instructions are stored and the processing resource for executing the instructions, or the machine-readable storage medium may be separate but accessible to the system 400 and the processing resource. In other examples, the processing engine(s) 508 may be implemented by electronic circuitry.

[0090] The database 510 may include data that is either stored or generated as a result of functionality implemented by one of the components of the processing engine(s) 508. In one embodiment, the processing engine(s) 508 include a signal triggering unit 512, an activation unit 514, a sending unit 516, and other unit(s) 518. The other unit(s) 518 may implement functionality that complements applications / functions performed by the control unit 406.

[0091] In one aspect, the signal triggering unit 512 triggers a first signal to activate the spectrometer 402 and / or the system 400 to receive and analyze the input spectra comprising real-time spectrometric measurement data of the battery cell and to generate the quality status of the battery cell based on the analyzed input spectra.

[0092] In an exemplary embodiment, the control unit 406 triggers a second signal, and upon receipt of the second signal, the activation unit 514 then assists in deactivating the spectrometer 402 and / or the system 400. The activation unit 514 enables selective activation and deactivation of the spectrometer 402 and / or the system 400 based on the detected first signal by the control unit 406. The activation unit 514 may also enable deactivation of the spectrometer 402 when not in use.

[0093] According to an exemplary embodiment, the transmitting unit 516 supports the transmission of the signals associated with the spectrometer 402.

[0094] Furthermore, in an embodiment of the present disclosure, the image recognition module may employ any image recognition technique, including, but not limited to, a deep learning technique or model such as a convolutional neural network (CNN) technique, etc., an artificial intelligence (AI) technique, a machine learning (ML) technique, or the like.

[0095] Now, in a non-limiting and exemplary embodiment of the present disclosure, with reference to Fig. 6A and Fig. 6B illustrates the operation of an image recognition module of the proposed system using a convolutional neural network (CNN) technique according to an embodiment of the present disclosure.

[0096] One of the convolutional neural network techniques, such as CNN-LeNet-5 or the like, can be used here. CNN-LeNet-5 can consist of applying convolutional layers to understand the hierarchical pattern in data, which is a significant departure from fully connected layers (FCLs). It can be designed using several key ideas, including: local receptive fields, where each neuron in a layer is connected to only a small region of the previous layer, reducing the number of parameters and allowing the CNN to learn local features useful for the task at hand; and weight sharing, where each neuron in a given feature map shares the same set of weights with all other neurons in that feature map.

[0097] This sharing of weights further reduces the number of parameters and favors the CNN to learn features that are invariant to small translations of the input data / input spectra.

[0098] It also includes subsampling, which uses subsampling layers to reduce the spatial size of feature maps while retaining important information, reducing the computational burden of the CNN and helping prevent overfitting; and convolutional layers, which allow the network to learn spatially invariant features useful for the task at hand. This is especially important for deep learning tasks, such as input spectra, where the Raman spectra of the battery cell in the image may vary.

[0099] The CNN-LeNet architecture can consist of two convolutional layers, two subsampling layers (max-pooling layer) and three fully connected layers, as shown in Fig. 6A. Overall, the CNN / CNN-LeNet architecture demonstrates the power of convolutional neural networks for image recognition tasks.

[0100] As a non-limiting example, consider an input image with height H, width W, and C channels (for example, an RGB image has 3 channels). There is also a set of K learnable filters (kernels), where each filter has a height, width, and channels. The channels are typically significantly larger than the input number of channels, and the filter height and width are significantly smaller than the spatial resolution of the input.

[0101] The 2D convolution process takes each filter and slides it over the input image. At each position, an element-wise multiplication is performed between the filter and the corresponding pixels of the input image, and then the resulting values ​​are summed to produce a single output value.

[0102] This operation corresponds to a matrix multiplication operation with a matrix weight composed of the repeated kernel values, making it equivalent to a fully connected weight-sharing layer. It is also important to note that the pixel values ​​of an input image / spectrum are not exactly connected to the output layer, but rather in partially connected layers.

[0103] To compute the output feature map for all filters, the convolution process can be performed K times, once for each filter. The output feature map can then be obtained by stacking the feature maps along the depth axis. Here, the size of the output feature map is determined by the size of the input image / spectra, the filter size, the step size, and the padding.

[0104] Furthermore, with reference to Fig. Figure 6B illustrates the operation of the CNN module according to the proposed spectrum recognition method 100. It consists of a number of convolutional layers for feature extraction and two fully connected layers for classifying the spectrum(s) of the input spectra.

[0105] With reference to Fig. 7A and Fig.7B, exemplary spectra of a faulty battery cell and an intact battery cell, as each is detected by the proposed system, are illustrated according to an exemplary embodiment of the present disclosure.

[0106] The spectra 700a for a faulty battery cell illustrate an increase in Raman shift distortion with increasing Raman spectrum (RS) intensity. For example, from 3.86 au (arbitrary units) to 3.935 au, there is no change in the Raman shift, while above 3.935 au to 4.2 au, a significant Raman shift is evident in the spectrum.

[0107] As can be seen from spectra 700b, no change in the Raman shift occurs with increasing Raman spectrum (RS) intensity for an integrated battery cell. For example, when the RS intensity changes from 3.20 au itself to 4.45 au, no explicit change / variation in the Raman shift occurs for the integrated battery cell.

[0108] While various embodiments of the invention have been described above, other and further embodiments of the invention may be devised without departing from the basic scope thereof. The scope of the invention is determined by the following claims. The invention is not limited to the described embodiments, versions, or examples. These are provided here to enable one of ordinary skill in the art to make and use the invention in combination with information and knowledge available to those skilled in the art.

[0109] The present disclosure provides a method and system that overcomes the above-mentioned limitations of existing systems and / or methods for determining the quality of battery cells using spectroscopy.

[0110] The present disclosure provides a method and system for determining the quality of the battery cell in the assembly line prior to installation of the battery cell in the vehicle.

[0111] The present disclosure provides a method and system for determining the quality of the battery cell in the assembly line using spectrometric techniques, Raman spectrum, which is used for predicting battery cell failure under steady-state conditions.

[0112] The present disclosure provides a method and system for determining battery cell quality in the assembly line that supports the creation of a digital passport for each battery cell or battery pack.

[0113] The present disclosure provides a method and system for determining the quality of the battery cell in the assembly line to evaluate the reuse of the battery cell in other applications (second life).

[0114] The present disclosure provides a method and system for determining battery cell quality in the assembly line for accurate detection of battery cell failures.

[0115] The present disclosure provides a method and system for determining battery cell quality in the assembly line that reduces unplanned vehicle downtime by early detection of battery failures.

[0116] The present disclosure provides a method and system for determining battery cell quality in the assembly line that supports root cause analysis of aging / failures using explainable artificial intelligence (AI) or machine learning (ML) techniques. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] US 10502793B2

[0006]

Claims

[1] Method (100) for determining the quality of a battery cell, the method (100) comprising the steps of: Receiving (102), by a processor (502) of a control unit from a spectrometer (402), an input Raman spectrum of the battery cell (406) representing intensity values ​​for a range of wavelengths; Comparing (104), by the processor (502), at least one feature extracted from the input Raman spectra with at least one model feature of a quality determination module and based on the comparison, determining (106), by the processor (502), whether the integrity state of the battery cell is faulty or fault-free. [2] The method (100) of claim 1, wherein the at least one feature comprises intensity values ​​at one or more wavelengths and the at least one model feature comprises one or more identified wavelengths and weights assigned to the one or more identified wavelengths. [3] The method of any one of claims 1 to 2, wherein the battery cell is determined to be the faulty cell if the intensity values ​​at the one or more identified wavelengths are greater than a threshold value. [4] The method (100) of claim 1, wherein the method (100) for determining the quality of the battery cell is performed prior to installation of the battery cell in a vehicle. [5] The method (100) of claim 1, wherein the quality determination module is obtained by: Collecting a Raman spectrum data item from a plurality of example battery cells by the processor (502) using the spectrometer (402), wherein the plurality of example battery cells comprises at least one of defective battery cells, non-defective battery cells, and at least one used battery cell; performing a baseline correction, a smoothing treatment, and a normalization treatment on the Raman spectrum data item by the processor (502) to obtain preprocessed spectrum data; and Obtaining the quality determination module by the processor (502) by taking the preprocessed spectrum data, the step of obtaining comprising: Assigning all Raman spectra from preprocessed spectrum data to the integrity state of the corresponding Raman spectra; Splitting the preprocessed spectrum data into a training spectrum data item and a test spectrum data item and Input the training spectrum data item and the test spectrum data item into a machine learning model to obtain the quality determination module. [6] The method (100) of claim 5, wherein the step of obtaining the quality determination module comprises: training the machine learning model, wherein training includes: analyzing the training spectrum data item to identify the at least one model feature in the training spectrum data item. [7] Method according to one of the preceding claims 5 to 6, wherein the machine learning model is a convolutional neural network (CNN). [8] System (400) for determining the quality of a battery cell, the system (400) comprising: a spectrometer (402) and a control unit (406) communicatively connected to the spectrometer (402), the control unit (406) comprising a processor (502) coupled to a memory (504), the memory (504) storing one or more instructions executable by the processor (502) for: Receiving, from the spectrometer (402), an input Raman spectrum of the battery cell representing intensity values ​​for a range of wavelengths; Comparing at least one feature extracted from the input Raman spectra with at least one model feature of a quality determination module and Based on the comparison, determine whether the battery cell integrity state is faulty or fault-free. [9] The system (400) of claim 8, wherein the control unit (406) is configured, in case of receiving the quality determination module, to: Collecting a Raman spectrum data item from a plurality of example battery cells by the processor (502) using the spectrometer (402), wherein the plurality of example battery cells comprise at least one of defective battery cells, non-defective battery cells, and at least one used battery cell; performing a baseline correction, a smoothing treatment, and a normalization treatment on the Raman spectrum data item by the processor (502) to obtain preprocessed spectrum data; and Obtaining the quality determination module by the processor (502) by taking the preprocessed spectrum data, wherein the control unit (406) for obtaining the quality determination module is further configured to: Assigning all Raman spectra from preprocessed spectrum data to the integrity state of the corresponding Raman spectra; Splitting the preprocessed spectrum data into a training spectrum data item and a test spectrum data item and Input the training spectrum data item and the test spectrum data item into a machine learning model to obtain the quality determination module. [10] System (400) according to one of the preceding claims 8 to 9, wherein: the at least one feature comprises intensity values ​​at one or more wavelengths and the at least one model feature comprises one or comprises a plurality of identified wavelengths and weights assigned to the one or more identified wavelengths; the battery cell is determined to be the faulty cell if the intensity values ​​at the one or more identified wavelengths are greater than a threshold value; the control unit (406) is configured to determine the quality of the battery cell prior to installation of the battery cell in a vehicle; and the machine learning model is a convolutional neural network (CNN).

Images