SYSTEM FOR ESTIMATING DISCHARGE CAPACITY OF LITHIUM- ION CELL AND METHOD THEREOF

Abstract

The disclosure presents a system (200) and method (300) for rapid discharge capacity estimation of lithium-ion cells. The system (200) and method (300) involves charging the cell at a specified C-rate in Constant Current (CC) mode and collecting the charging data. The collected data is transmitted to a pre-trained machine-learning model. A voltage sensor (208) detects the electrical potential difference during charging, and the data is sent to processors for analysis. Key features, including voltage values at equidistant intervals, derivatives of these values, and differences between these derivatives and a reference curve, are extracted. The extracted features are scaled to normalize the data and then input into the machine-learning model to predict the cell’s discharge capacity. The system (200) and method (300) significantly reduces the time required for discharge capacity estimation, providing accurate results within 15 minutes, with an error margin of less than or equal to 5%.

Specification

Description:TECHNICAL FIELD
[0001] The present disclosure pertains to the field of lithium-ion cell technology. More specifically, it relates to a system and method designed for estimating the discharge capacity of lithium-ion cell.

BACKGROUND
[0002] Background description includes information that may be useful in understanding the present disclosure. It is not an admission that any of the information provided herein is prior art or relevant to the presently claimed disclosure, or that any publication specifically or implicitly referenced is prior art.
[0003] Lithium-ion cell, widely used in a range of applications, undergo degradation over time and with repeated cycles. This degradation is reflected in changes to their voltage profiles, which contain critical information about the cell's discharge capacity. One of the key phenomena indicating this degradation is known as 'voltage drift'. Voltage drift refers to the gradual change in the voltage characteristics of a cell as it ages, and the extent of this drift can provide valuable insights into the cell's condition. Additionally, another important indicator is the rate of change of voltage with respect to time, denoted as dV/dt. This derivative reflects how quickly the voltage changes during charging and discharging cycles, and deviations in this rate can signal changes in the cell's discharge capacity.
[0004] Conventional method in assessing discharge capacity involves performing a full charge-discharge and measuring the capacity discharged during a complete discharge. However, the process is time-consuming and may not be necessary for accurate discharge capacity estimation.
[0005] FIG. 1 illustrates an exemplary graphical representation of voltage during conventional lithium-ion cell testing, in accordance with an embodiment of the present disclosure.
[0006] In an exemplary embodiment, referring to FIG. 1, the file "voltage_conventional_test.png" features a plot titled "Voltage during conventional testing." The graphical representation 100 can illustrate the variation in voltage during the conventional cell testing process used to estimate discharge capacity. The plot identifies different steps in the process: CC (Constant Current) Charging - 1, CV (Constant Voltage) Charging - 2, Rest - 3, CC Discharging - 4, and CC Charging - 5. The total testing time is approximately 6.6 hours.
[0007] Currently, the assessment of the discharge capacity of lithium-ion cell is a critical yet challenging task, with only a few viable technologies available that can perform this function quickly and accurately. The standard industry practice involves subjecting each cell to a complete charge-discharge cycle under specific conditions prescribed by the manufacturer to determine its discharge capacity. While this method is effective, it is also highly time-consuming, often requiring up to five hours to yield results. This lengthy process is not only labor-intensive but also inefficient, consuming valuable man-hours and resources that could be better utilized elsewhere. The need for a quicker and more efficient method is evident, particularly in industries where fast cell diagnostics are crucial for maintaining operational efficiency and safety.
[0008] There is, therefore, a need to address these challenges by introducing a machine learning approach for estimating the discharge capacity of the lithium-ion cells.

OBJECTS OF THE PRESENT DISCLOSURE
[0009] Some of the objects of the present disclosure, which at least one embodiment herein satisfies are as listed herein below.
[0010] An object of the present disclosure is to provide a method for discharge capacity estimation of lithium-ion cells that significantly reduces the testing time compared to traditional methods, allowing for quick and efficient diagnostics.
[0011] Another object of the present disclosure is to utilize random and partial charging voltage curves for discharge capacity estimation.
[0012] Another object of the present disclosure is to leverage the phenomenon of voltage drift and the rate of change of voltage (dV/dt) as key indicators for assessing cell discharge capacity.
[0013] Another object of the present disclosure is to integrate advanced machine learning algorithms with the proposed method, enhancing the accuracy of discharge capacity estimation through sophisticated data analysis and pattern recognition techniques.
[0014] Another object of the present disclosure is to ensure the method is adaptable to various types of lithium-ion cell used in different applications.
[0015] Another object of the present disclosure is to facilitate the early detection of lithium-ion cell degradation and potential failures, enabling proactive maintenance and replacement strategies to prevent unexpected downtime and extend cell life.

SUMMARY
[0016] Various aspects of the present disclosure pertain to the field of lithium-ion cell technology. More specifically, it relates to a system and method designed for estimating the discharge capacity of lithium-ion cells.
[0017] An aspect of the present disclosure pertains a system for discharge capacity estimation of the lithium-ion cell, where the system may be configured to include a voltage sensor coupled with an Analog to Digital Converter (ADC) configured to detect the electrical potential difference across two points of the lithium-ion cell; where the ADC sample analog voltage signals into corresponding digital values; and a server operatively coupled to the voltage sensor, the server including one or more processors coupled with a memory, the memory storing instructions executable by the one or more processors configured to receive the detected voltage data for estimating the input charging data and the observed voltage drift; extract the received voltage values at equidistant intervals from an original voltage curve recorded during the charging process; addition square of derivative, cosine transformation, and constants of the extracted voltage values at one or more selected equidistant points; normalize the squared difference in voltage derivatives, and magnify the changes between the input's derivative and the reference's derivative; scale the determined difference between the derivatives of the input voltage curve and the reference voltage curve to normalize the processed data; and input the scaled features into the machine learning model to predict the discharge capacity of the lithium-ion cell.
[0018] In an aspect, the one or more processors configured to perform pre-processing of the digital values, including noise reduction and signal smoothing, before storing the data in the memory.
[0019] Another aspect of the present disclosure pertains to a method for discharge capacity estimation of a lithium-ion cell may be configured to include several steps. First, the lithium-ion cell is charged at a C-rate using the Constant Current (CC) mode. During this process, the charging data obtained is collected and transmitted to a machine-learning model, which has been pre-trained on the charging data of the lithium-ion cell.
[0020] Concurrently, the electrical potential difference across two points of the lithium-ion cell is detected via a voltage sensor. The detected voltage data is then transmitted to one or more processors for estimating the input charging data and observing the voltage drift, leading to the termination of the charging process after a predefined period. The method can continue with the extraction, by the processors, of the features from the transmitted data.
[0021] Furthermore, the extraction process includes extracting the transmitted voltage values at equidistant intervals from an original voltage curve recorded during the charging process; adding, by the one or more processor, a square of derivative, cosine transformation, and constants of the extracted voltage values at one or more selected equidistant points; and normalizing, by the one or more processor, the squared difference in voltage derivatives, and magnifying the changes between the input's derivative and the reference's derivative; wherein the reference voltage curve configured to obtain at the same voltage cell conditions as that of the actual data.
[0022] In an aspect, the method for managing the data collected from the charging voltage curves of lithium-ion cells includes identifying and removing, noise, errors, and irrelevant information from the voltage dataset; re-sampling the sensed voltage data by applying interpolation techniques and piece-wise linear functions to adjust the sensed voltage data points, configured to ensure that the data is uniformly sampled for consistent analysis; extracting, the relevant features from the re-sampled data, identifying key characteristics and patterns for accurate discharge capacity estimation; and scaling, the extracted features using predefined scaling artifacts to normalize the data, ensuring that the extracted features are on a comparable scale and suitable for input into the machine learning model.
[0023] In an aspect, the extraction of the equidistant voltage values can involve identifying and sampling, by the one or more processor, the voltage measurements at regular and uniform intervals to create a dataset of the extracted equidistant voltage values.
[0024] In an aspect, the derivative of the extracted voltage may involve computing, by the one or more processor, the rate of change of the extracted voltage over time at the one or more of the selected equidistant points to capture the dynamic behaviour of the extracted voltage during the charging process.
[0025] In an aspect, the machine-learning model including, any or a combination of, a neural network, and predictive model trained on a large dataset of charging cycles from the cells.
[0026] In one aspect, the machine-learning model including, a neural network, and predictive model trained on a large dataset of charging cycles from the lithium-ion cell.
[0027] In one aspect, the voltage drift refers to a gradual change in voltage characteristics of the cell during repeated charging cycles.
[0028] In one aspect, the charging process configured to monitor in real-time, and the charging data is continuously fed to the machine-learning model for dynamic discharge capacity estimation.
[0029] Various objects, features, aspects, and advantages of the inventive subject matter will become more apparent from the following detailed description of preferred embodiments, along with the accompanying drawing figures in which numerals represent like components.

BRIEF DESCRIPTION OF DRAWINGS
[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.
[0031] FIG. 1 illustrates an exemplary graphical representation of voltage during conventional lithium-ion cell testing, in accordance with an embodiment of the present disclosure.
[0032] FIG. 2 illustrates an exemplary architecture for discharge capacity estimation of lithium-ion cell, in accordance with an embodiment of the present disclosure.
[0033] FIG. 3 illustrates an exemplary view of a flow diagram of proposed method for discharge capacity estimation of lithium-ion cell, in accordance with an embodiment of the present disclosure.
[0034] FIG. 4 illustrates an exemplary graphical representation of test results of the system and method on cells, with predictions made in units of discharge capacity, in accordance with an embodiment of the present disclosure.

DETAILED DESCRIPTION
[0035] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such detail as to clearly communicate the disclosure. However, the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure as defined by the appended claims.
[0036] The proposed disclosure pertains to a discharge capacity and efficient way to estimate of the lithium-ion cell, even when the cell's history is unknown, within just 15 minutes. This process begins by charging the provided cell at the manufacturer-specified C-rate, ensuring standardized charging conditions. The charging data is then fed into a machine-learning model that has been pre-trained on extensive datasets of charging profiles from numerous cells of the same make. A critical component of this model is its ability to analyse the phenomenon known as 'voltage drift', which refers to the gradual changes in a cell's voltage profile over time due to aging and usage. By examining these changes, the model can accurately infer the cell's degradation and predict its discharge capacity. This approach achieves a high level of accuracy, with predictions made within an error margin of less than or equal to 5%, providing a reliable and quick assessment of cell discharge capacity.
[0037] An embodiment of the present disclosure a system for discharge capacity estimation of a lithium-ion cell, where the system may be configured to include a voltage sensor coupled with an Analog to Digital Converter (ADC) configured to detect the electrical potential difference across two points of the lithium-ion cell, where the ADC samples analog voltage signals into corresponding digital values.
[0038] Furthermore, the system can include a server operatively coupled to said voltage sensor, the server including one or more processors coupled with a memory, the memory storing instructions executable by the one or more processors configured to receive the detected voltage data for estimating the input charging data and the observed voltage drift; extract the received voltage values at equidistant intervals from an original voltage curve recorded during the charging process; addition square of derivative, cosine transformation, and constants of the extracted voltage values at one or more selected equidistant points; normalize the squared difference in voltage derivatives, and magnify the changes between the input's derivative and the reference's derivative; scale the determined difference between the derivatives of the input voltage curve and the reference voltage curve to normalize the processed data; and input the scaled features into the machine learning model to predict the discharge capacity of the lithium-ion cell.
[0039] An embodiment of the present disclosure a method for estimating the discharge capacity of a lithium-ion cell involves charging the cell at a C-rate using the Constant Current (CC) mode and collecting the charging data, which is then transmitted to a pre-trained machine-learning model. During charging, the voltage difference across two points of the cell is detected by a voltage sensor and transmitted to a processor to estimate charging data and observe voltage drift, leading to the termination of the charging process after a set period. The processor then extracts features from the data by taking voltage values at equidistant intervals from the original voltage curve, adding a square of derivative, cosine transformation, and constants of the extracted voltage values at one or more selected equidistant points; and normalizing the squared difference in voltage derivatives, and magnifies the changes between the input's derivative and the reference's derivative, where the reference voltage curve configured to obtain at the same voltage cell conditions as that of the actual data.
[0040] The extracted features are scaled to normalize the data and then input into the machine-learning model to predict the cell's discharge capacity.
[0041] FIG. 2 illustrates an exemplary architecture for discharge capacity estimation of lithium-ion cell, in accordance with an embodiment of the present disclosure.
[0042] In an embodiment, referring to FIG. 2, a system 200 may be configured to design for discharge capacity estimation of lithium-ion cell (interchangeably referred to as a cell, hereinafter). The system 200 can include several key components working in tandem to deliver precise and discharge capacity assessments. The system 200 can include a voltage sensor 208 may be configured to detect and measure an electrical potential difference, or voltage, across two points in an electrical circuit of the system 200.
[0043] In an exemplary embodiment, the voltage sensor 208 may be monitoring the voltage levels during one or more stages of the cell's charge and discharge cycles. The voltage sensor 208 can operate by converting the analog voltage signal into a form that can be processed digitally, typically through an Analog to Digital Converter (ADC) 210. The conversion is essential for precise data analysis and subsequent processing by the system's micro-controller.
[0044] Furthermore, the accuracy of the voltage sensor 208 is vital, as even slight deviations can impact the reliability of the discharge capacity predictions. To ensure compatibility with one or more cell types and conditions, the voltage sensor 208 may be designed to handle a wide range of voltage levels, maintaining high resolution and accuracy. The capability can allow the voltage sensor 208 to capture detailed voltage profiles, which are integral to detecting phenomena such as voltage drift-a key indicator of cell degradation. By providing real-time, high-fidelity voltage data, the voltage sensor 208 can enable advanced analytical models to perform sophisticated assessments, ensuring that the discharge capacity of the cell can be determined quickly and accurately.
[0045] The system 200 can include the Analog to Digital Converter (ADC) 210 coupled with the voltage sensor 208, involved in monitoring and analysing cell discharge capacity. The ADC 210 is to translate the continuous, analog voltage signals detected by the voltage sensor 208 into discrete digital values that can be processed by digital systems, as including microcontrollers or computers.
[0046] In operation, the ADC 210 is configured to sample the analog voltage signal at a predetermined rate, known as the sampling rate. The rate is crucial because it determines how frequently the analog signal is measured, impacting the accuracy and resolution of the digital representation. A higher sampling rate captures more data points from the analog signal, leading to a more accurate and detailed digital representation.
[0047] In an exemplary embodiment, the system 200 may be configured to include a server 202 where the server 202 may be configured to include one or more processors 204 (interchangeably referred to as a processor 204, hereinafter) and a memory 206 storing a set of instructions, which upon being executed cause the processor 204. The processor 204 includes any or a combination of suitable logic, circuitry, and/or interfaces that are operable to execute one or more instructions stored in the memory 206 to perform pre-determined operations.
[0048] Furthermore, the processor 204 and memory 206 are integral components of the system 200 may be configured to design for estimating the discharge capacity of the lithium-ion cells. The processor 204, often a microcontroller, acts as the brain of the system 200, coordinating the one or more functions and processing the data collected by the voltage sensor 208 and converted by the Analog to Digital Converter (ADC) 210. The processor 204 may be responsible for executing the machine-learning algorithms that analyse the cell's voltage data to estimate its discharge capacity.
[0049] In an exemplary embodiment, the memory 206 can store the machine-learning model and the instructions necessary for data processing and analysis, and also holds the collected data from the voltage sensor 208, allowing for both real-time processing and historical data comparison. The memory 206 can include different types of storage, such as volatile memory (RAM) for temporary data storage during active computations and non-volatile memory (flash or EEPROM) for long-term storage of the machine-learning model and processed data.
[0050] Together, the processor 204 and memory 206 form the core computational engine of the system 200. The processor's 204 speed and efficiency determine how quickly and accurately the discharge capacity estimation can be performed, while the memory's 206 capacity and reliability ensure that all necessary data and instructions are readily available for processing. The synergy between the processor 204 and memory 206 can enable the system 200 to deliver fast and precise cell discharge capacity assessments, making it a powerful tool for maintaining and managing lithium-ion cell.
[0051] The processor 204 may be implemented using one or more processors technologies known in the art. Examples of the processor 204 include but are not limited to, a microcontroller, an x86 processor, a RISC processor, an ASIC processor, a CISC processor, or any other processor.
[0052] In an exemplary embodiment, the system 200 may be configured to include a display unit 212 may be configured to design for estimating the discharge capacity of the lithium-ion cells. The display unit 212 typically features a Liquid Crystal Display (LCD) screen, which can serve as the interface between the interface and a user (not shown). The primary function of the display unit is to present the output of the machine-learning model's inference in a clear and accessible manner. After the processor 204 analyses the voltage data and estimates the discharge capacity, the results may be relayed to the display unit 212.
[0053] The LCD screen may be chosen for its readability and efficiency, providing a user-friendly platform for displaying vital information including, any or a combination of, but not limited to, the estimated discharge capacity, voltage readings, and any one or more relevant data or alerts. The display unit 212 can ensure that the user can quickly and easily interpret the results of the cell discharge capacity assessment, making informed decisions based on accurate and real-time data.
[0054] Additionally, the display unit 212 may be equipped with features to enhance the user interaction, including touch functionality for navigating through different screens or settings. The display unit 212 can also show graphical representations of the voltage profiles and other diagnostic information, offering a comprehensive view of the cell's condition. By integrating the display unit 212, the system 200 can become a complete and intuitive tool for monitoring and managing the discharge capacity of lithium-ion cell.
[0055] In an exemplary embodiment, the system 200 may be configured to include a power unit 214 may be designed for estimating the discharge capacity of the lithium-ion cell. This power unit 214 can responsible for supplying the necessary energy to all one or more components of the system 200, ensuring consistent and reliable operation. The power unit 214 typically houses an internal cell, which allows the system 200 to function independently without requiring an external power source.
[0056] The internal cell is carefully selected to provide sufficient capacity to power the voltage sensors 208, the Analog to Digital Converter (ADC) 210, the processor 204, the memory 206, and the display unit 212 throughout the discharge capacity estimation process. The power unit 214 may ensure that the device can be used in various environments and conditions, offering flexibility and portability for the users who need to perform cell discharge capacity assessments in the field or on-site.
[0057] Moreover, the power unit 214 often includes circuitry to manage the charging and discharging of the internal cell, optimizing its lifespan and maintaining the system's operational efficiency. By providing a stable and reliable power supply, the power unit 214 can play a critical role in the overall functionality and usability of the system 200, enabling accurate and discharge capacity estimation for lithium-ion cell.
[0058] In an exemplary embodiment, the system 200 may be configured to include a network (not shown) designed for estimating the discharge capacity of the lithium-ion cells, facilitating seamless communication and data transfer between the device and external systems. The network connectivity can be achieved through various means such as Wi-Fi, Bluetooth, or cellular networks, depending on the design and application requirements.
[0059] By integrating network capabilities, the system 200 can transmit the collected voltage data and discharge capacity estimation results to remote servers or cloud-based platforms in real-time. The connectivity enables remote monitoring and analysis, allowing the users to access the cell discharge capacity information from anywhere, without being physically present near the device. Furthermore, the network integration supports software updates and remote diagnostics, ensuring the device remains up-to-date with the latest advancements and can be troubleshooted or enhanced without direct intervention.
[0060] Additionally, the network may also allow for the aggregation of data from multiple devices, providing a broader dataset for machine learning models, which can improve the accuracy and reliability of discharge capacity predictions. In industrial or large-scale cell management scenarios, the capability is crucial for maintaining an overview of the discharge capacity and performance of numerous cell simultaneously. Overall, the network component enhances the system's functionality, offering robust, flexible, and scalable solutions for cell discharge capacity management.
[0061] In an exemplary embodiment, data collection may be conducted at room temperature utilizing the SEMCO 5V-10A cylindrical cell testing equipment. The specific parameter values for each step in the testing protocol can meticulously determine based on the detailed specifications provided in the cell datasheet prepared by the cell manufacturer. The datasheet outlines the optimal conditions and operational limits for the cell, ensuring that the testing protocol aligns with the manufacturer's guidelines to obtain accurate and reliable data. By adhering to these manufacturer-recommended parameters, the integrity and consistency of the data collected during the testing process are maintained, providing a robust foundation for subsequent analysis and evaluation of the lithium-ion cell's performance.
[0062] In an exemplary embodiment, re-sampling using interpolation and piece-wise linear functions: In an exemplary embodiment, given a set of data points (v1, t1), (v2, t2), . . . , (vn, tn), which are present in the original voltage snippet, interpolate and subsequently extract 30 equidistant points from the curve at 30-second intervals. The formula for linear interpolation between two points (vi, ti) and (vi+1, ti+1) is:

v=v_i+ (v_(i+1)-v_i)/(t_(i+1)-t_i ) (t-t_i)
where ti = t = ti+1.
The interpolated voltage curve is given by V(t), a series of piece-wise linear functions as described, below;
V(t)={¦(v_0+ (v_1-v_0)/(t_1-t_0 ) (t-t_0 ),t_0 = t = t_1@v_1+ (v_2-v_1)/(t_2-t_1 ) (t-t_1 ),t_1 = t = t_2@.@.@.@v_(n-1)+ (v_n-v_(n-1))/(t_n-t_(n-1) ) (t-t_(n-1) ),t_(n-1) = t = t_n )¦
Once the function has been defined, proceed to extract 30 equidistant points from the voltage curve adding up to the first 900 seconds. Therefore, the final extracted snippet is given by, Vinput = V(t) where t ? {30, 60, 90, . . . , 900}.
Feature extraction: three features are provided as inputs to the model across 30 time-steps. These features are:
1. Voltage (Vinput): the equidistant voltage values extracted from the original curve
2. a_2 (dV/dt)^2 + b_2 cos(dV/dt) + c_2
term 1 i.e. square of the derivative amplifies significant changes in voltage over time.
term 2 i.e. cosine transformation smooths smaller fluctuations and captures cyclical patterns in dV/dt.
values of the coefficients and constants are:
a2 = 1.2
b2 = 0.85 if 2.5 < V < 3.3; 0.16 if 3.3 < V < 4.1
c2 = 0.55
these coefficients and constants are cell model specific and were experimentally determined and optimized during training via gradient descent.
3. a3(dVinput /dt - dVreference/dt) / V + exp(a3(dVinput /dt - dVreference/dt)) + b3
term 1 normalizes the squared difference in voltage derivatives to ensure that the magnitude of the difference is weighed appropriately for different voltage levels.
term 2 is employed to magnify the changes between the input's derivative and the reference's derivative. This is particularly important when the deviations are very small, but could indicate early signs of battery degradation.
a3 = 0.95 if 2.5 < V < 3.3; 0.11 if 3.3 < V < 4.1
b3 = 0.08
The third feature works perfectly within the limitations of the problem as the derivative of the voltage at each point can be connected to the voltage itself. This stems from the characteristics of the voltage profile itself:
The mapping between V and t is injective (one to one) on the set [2.5, 4.1].
?t1, t2 ? [2.5, 4.1], t1 ? t2 =? V(t1) ? V(t2)
Or more concisely:
V is injective on [2.5, 4.1]
2. Naturally, the derivative of the inverse of the aforementioned functions,dt/dV, will also have the same predictive power and may be used as the third feature instead of the derivative of Voltage.
Feature scaling: once the feature extraction has been completed, perform scaling to ensure that the inputs provided to the model are consistent and to avoid large values. Each feature is scaled separately and the scaling parameters are then stored for later use during inference. Standard scaling, also known as z-score normalization, is a technique used to standardize the features of a dataset. The process transforms the data such that the distribution of features has a mean of zero and a standard deviation of one.
The standardized value z of a feature x is computed as follows:
z=(x-µ)/s
where:
• x is the original value of the feature,
• µ is the mean of the feature,
• s is the standard deviation of the feature.
The mean µ and standard deviation s are calculated as:
µ=1/N ?_(i=1)^N¦x_i
s=v(1/N ?_(i=1)^N¦?(x?_i ?-µ)?^2 )
where N is the number of samples, and xi represents each individual sample of the feature. Standard scaling ensures that the transformed data follows a standard normal distribution.
[0066] In an exemplary embodiment, Convolutional Neural Networks (CNNs) may be employed in the model to capture subtle variations in the undulations and gradients of the voltage profile. While CNNs are traditionally utilized for image processing and segmentation tasks in computer vision problems, they are also effective for extracting features from time-series data. This latter characteristic of CNNs makes them particularly well-suited for the purposes in analysing voltage profiles.
[0067] In an exemplary embodiment, three major types of layers are used in the model, namely,
1. 1D Convolutional layers: Layers apply convolution operations to one-dimensional data, such as time-series data. In this context, they are used to detect local patterns and subtle variations within the voltage profile, such as changes in the undulations and gradients. The 1D convolutional layers help in capturing these features by applying filters across the data to extract relevant information that is crucial for the model's predictions.
2. Pooling layers: Pooling layers may be used to reduce the dimensionality of the data, which helps in minimizing the computational load and preventing overfitting. In the model, pooling layers down-sample the output from the convolutional layers by summarizing the presence of features in certain regions. This is typically done using operations like max pooling or average pooling, which retain the most significant information while discarding less relevant details.
[0068] Dense layers: Also known as fully connected layers, dense layers connect every neuron in one layer to every neuron in the next layer. These layers are used to integrate the features extracted by the convolutional and pooling layers and to perform the final classification or regression tasks. In the model, dense layers are responsible for interpreting the extracted features and making predictions about the discharge capacity of the lithium-ion cell based on the processed voltage profile data.
[0069] FIG. 3 illustrates an exemplary view of a flow diagram of proposed method for discharge capacity estimation of lithium-ion cell, in accordance with an embodiment of the present disclosure.
[0070] As illustrated, a method 300 for fast way to estimate the discharge capacity of a lithium-ion cell with unknown history within 15 minutes. At step 302, the method 300 may involve connecting the used cell to a charging apparatus. The used cell may be plugged into the charging equipment, which is configured to commence the charging process. The charging equipment operates according to the manufacturer-specified C-rate, ensuring that the cell is charged under optimal and controlled conditions.
[0071] Continuing further, at step 304, the method 300 may involve collecting the charging data obtained during the cell's charging process and transmitting the data to a machine-learning model. The machine-learning model may be pre-trained on historical charging data from similar cell cells.
[0072] Continuing further, at step 304-1, the method 300 may involve connecting the sensing wires to the terminals of the cell. The sensing wires are securely attached to the positive and negative terminals of the cell, ensuring a stable and reliable electrical connection. The connection allows for accurate measurement of the voltage across the cell terminals, which is essential for collecting the data necessary for subsequent analysis and discharge capacity estimation.
[0073] Continuing further, at step 304-2, the method 300 may involve the step of initiating the charging process. The charging is commenced by activating the charging equipment, which begins to supply electrical current to the cell according to the manufacturer-specified C-rate. This controlled charging process is essential for collecting accurate and consistent voltage data, which may be used for subsequent analysis and discharge capacity estimation.
[0074] Continuing further, at step 306, the method 300 may involve utilizing a voltage sensor 208 to detect the voltage across the cell terminals. The voltage sensor 208 can measure the electrical potential difference and converts the analog voltage signal into corresponding digital values. These digital voltage values are then transmitted to one or more processor 204 (interchangeably referred to as a processor 204, hereinafter). The processor may be configured to process the data for further analysis, enabling accurate discharge capacity estimation of the lithium-ion cell.
[0075] At step 306-1 the method 300 may involve measuring the voltage difference between two points of the lithium-ion cell using the voltage sensor 208 during charging. Further, at step 306-2, the voltage data is then sent to the processor 204, which uses it to estimate the charging data and detect any voltage drift. Based on this information, the charging process is stopped after a set time period.
[0076] Continuing further, at step 308, the method 300 may involve terminating the charging process after a period of 15 minutes. The charging equipment, which is configured to charge the cell according to manufacturer-specified parameters, is programmed to cease supplying electrical current to the cell after the designated 15-minute interval.
[0077] Continuing further, at step 310-1, the method 300 may involve for data cleaning, where the processor 204 can identify and remove any noise, errors, or irrelevant information from the raw dataset. The cleaning process ensures that only accurate and pertinent data is retained for further analysis.
[0078] Continuing further, at step 310-2, the method 300 may involve applying interpolation techniques, such as linear interpolation or piece-wise linear functions for estimating values between existing data points. The step fills in gaps and smooths out the data to create a more uniformly sampled dataset. Furthermore, the method 300 may utilizing piece-wise linear functions to refine the dataset by adjusting segments between consecutive data points. This ensures that transitions between data points are gradual and consistent.
[0079] Continuing further, at step 310-3, the method 300 may involve analysing re-sampled data for discharge capacity estimation; the processor 204 may extract relevant features that capture essential characteristics and patterns. The extraction involves computational techniques designed to identify significant aspects of the re-sampled dataset, such as variations, trends, or anomalies that are indicative of cell discharge capacity. By applying algorithms tailored to feature extraction, the processor discerns key data attributes that contribute to accurate discharge capacity estimation. There may be three steps in the step 310-3 defines as; at step 310-3A, the processor 204 can extract voltage values at equidistant intervals from the original voltage curve recorded during the charging process.
[0080] Moreover, at step 310-3B, the processor 204 can add a square of derivative, cosine transformation, and constants of the extracted voltage values at one or more selected equidistant points, and at step 310-3C, the processor 204 can normalize 310-3C the squared difference in voltage derivatives, and magnify the changes between the input's derivative and the reference's derivative; wherein the reference voltage curve configured to obtain at the same voltage cell conditions as that of the actual data with the reference curve obtained under the same cell conditions as the actual data.
[0081] Continuing further, at step 310-4, the method 300 may involve applying predefined scaling artifacts or techniques that adjust the range or distribution of the extracted features. The normalization process aims to standardize the data, making features comparable in magnitude and suitable for effective integration into the machine learning model. By scaling the features, variations in their numerical ranges are minimized, thereby preventing biases that could affect the model's ability to interpret and learn from the data accurately.
[0082] Continuing further, at step 312-1, the method 300 may involve transferring processed data to a computational model. After processing raw data to extract relevant features and normalize those using predefined scaling methods, the processed data is then transmitted to the model. The step ensures that the model receives data that has been prepared and standardized for analysis. The model, which may include machine learning algorithms or analytical tools, utilizes the processed data to perform computations, generate predictions, or derive insights based on the trained parameters and data inputs.
[0083] Continuing further, at step 312-2, the method 300 may involve executing predictive operations by the model. After receiving processed data that has been prepared and standardized for analysis, the computational model utilizes its trained algorithms and parameters to perform predictions. These predictions are based on the processed data inputs, which may include normalized features extracted from raw data through pre-processing steps. The model applies statistical methods, machine learning techniques, or other analytical approaches to generate forecasts, classifications, or estimations relevant to the input data characteristics.
[0084] Continuing further, at step 312-3, the method 300 may involve the step of presenting prediction results derived from a computational model. After performing predictive operations using processed data inputs, the generated predictions are displayed on a display unit 212 integrated with the system 200. The display unit 212 can serve as a visual interface for presenting the outcomes of the computational model's analysis or decision-making processes.
[0085] FIG. 4 illustrates an exemplary graphical representation of test results of the system and method on cells, with predictions made in units of discharge capacity, in accordance with an embodiment of the present disclosure.
[0086] In an exemplary embodiment, illustrated in FIG. 4 the prediction test results 400 of the model on the test dataset. Further, k-fold cross validation may be performed to ensure the model's predictive capacity across all subsets of the dataset. The average evaluation metrics from this exercise is also provided in Table 1.


Table 1: Model evaluation results from 3 test datasets.
Evaluation test (Number of
cells) MAPE (Mean
Absolute Percentage
Error)
Test 1 (10568) 5.56%
Test 2 (2700) 4.07%
Test 3 (2150) 3.65%
[0087] In an exemplary embodiment, the evaluation tests 400 can provide insights into the accuracy of the predictions made by the methodology in relation to the true values of the cells. In Test 1, which involved 10,568 cells, the Mean Absolute Percentage Error (MAPE) is 5.56%, indicating that the predicted values are, on average, 5.56% different from the true values. Test 2, which included 2,700 cells, showed a reduced MAPE of 4.07%, demonstrating improved prediction accuracy. Test 3, with 2,150 cells, may be achieved the lowest MAPE of 3.65%, reflecting the highest accuracy among the tests. These results suggest that the prediction accuracy increases as the number of cells tested decreases, with lower MAPE values indicating closer alignment between the predicted and true values.
[0088] In summary, the present disclosure proposes an efficient system and method for estimating the discharge capacity of the lithium-ion cells within a significantly reduced timeframe of 15 minutes, as compared to conventional methods that may take several hours. The approach involves charging the cell at a manufacturer-specified rate and leveraging a pre-trained machine learning model that utilizes voltage drift phenomena. By analysing voltage profiles and extracting key features from the data, the model predicts discharge capacity with a high degree of accuracy, typically within an error margin of less than or equal to 5%. The system integrates voltage sensors, an Analog to Digital Converter (ADC), and a processor to collect and process data, ensuring precise estimation of cell discharge capacity. Results are displayed on a display unit for user accessibility, offering a speedy and reliable assessment tool for cell performance across various applications.
ADVANTAGES OF THE PRESENT DISCLOSURE
[0089] The proposed methodology can save over 95% of the time required for cell testing, allowing for faster project completion, quicker turnaround on testing results, and overall increased efficiency in the testing process.
[0090] The proposed methodology reduces the testing cost per cell by 19 times, making it highly cost-effective and enabling extensive testing within a constrained budget.
[0091] The proposed disclosure provides a method can effectively utilize low-frequency data, requiring only one data point per minute.
[0092] The proposed methodology can estimate the discharge capacity of used lithium-ion cells with any initial voltage between 2.5V and 4.1V.
[0093] The proposed disclosure completes the testing process in just 15 minutes. This significantly enhances efficiency, enabling quicker assessments and faster decision-making regarding the discharge capacity and usability of the lithium-ion cells.

, Claims:1. A system (200) for estimating discharge capacity of lithium-ion cell, wherein the system (200) comprises:
a voltage sensor (208) coupled with an Analog to Digital Converter (ADC) (210) configured to detect the electrical potential difference across two points of the lithium-ion cell; wherein the ADC (210) sample analog voltage signals into corresponding digital values; and
a server (202) operatively coupled to said voltage sensor (208), the server (102) comprising one or more processors (204) coupled with a memory (206), the memory (206) storing instructions executable by the one or more processors (204) configured to:
receive the detected voltage data for estimating the input charging data and the observed voltage drift;
extract the received voltage values at equidistant intervals from an original voltage curve recorded during the charging process;
addition square of derivative, cosine transformation, and constants of the extracted voltage values at one or more selected equidistant points;
normalize the squared difference in voltage derivatives, and magnify the changes between the input's derivative and the reference's derivative;
scale the determined difference between the derivatives of the input voltage curve and the reference voltage curve to normalize the processed data; and
input the scaled features into the machine learning model to predict the discharge capacity of the lithium-ion cell.

2. The system (200) as claimed in claim 1, wherein the one or more processor (204) configured to perform pre-processing of the digital values, comprising noise reduction and signal smoothing, before storing the data in the memory (206).

3. A method (300) for discharge capacity estimation of lithium-ion cell, comprising the steps of:
charging (302), the lithium-ion cell at a C-rate using the CC (Constant Current) mode;
collecting and transmitting (304), the charging data obtained from the charging process to a machine-learning model, wherein the machine-learning model configured to pre-train on the charging data of the lithium-ion cell;
detecting (306-1), the electrical potential difference across two points of the lithium-ion cell, via a voltage sensor (208), during the charging process; transmitting (306-2), the detected voltage data to one or more processor (204) for estimating the input charging data and the observed voltage drift, and correspondingly terminating (308), the charging process after a predefined period of time;
extracting (310), by the one or more processor (204), one or more features from the transmitted data; comprising the steps of:
extracting (310-3A), by the one or more processor (204), the transmitted voltage values at equidistant intervals from an original voltage curve recorded during the charging process;
adding (310-3B), by the one or more processor (204), square of derivative, cosine transformation, and constants of the extracted voltage values at one or more selected equidistant points; and
normalizing (310-3C), by the one or more processor (204), the squared difference in voltage derivatives, and magnifying the changes between the input's derivative and the reference's derivative; wherein the reference voltage curve configured to obtain at the same voltage cell conditions as that of the actual data.
scaling (310-4), by the one or more processor (204), the one or more extracted features to normalize the processed data; and
inputting (312), by the one or more processor (204), the scaled features into the machine learning model to predict the discharge capacity of the lithium-ion cell.

4. The method (300) as claimed in claim 3, wherein the method (310) for managing the data collected from the charging voltage curves of lithium-ion cells, comprising the steps of:
identifying and removing (310-1), by one or more processor (204), noise, errors, and irrelevant information from the voltage dataset;
re-sampling (310-2), by one or more processor (204), the sensed voltage data by applying interpolation techniques and piece-wise linear functions to adjust the sensed voltage data points, configured to ensure that the data is uniformly sampled for consistent analysis;
extracting (310-3), by the one or more processor (204), the relevant features from the re-sampled data, identifying key characteristics and patterns for accurate discharge capacity estimation; and
scaling (310-4), by the one or more processor (204), the extracted features using predefined scaling artifacts to normalize the data, ensuring that the extracted features are on a comparable scale and suitable for input into the machine learning model.

5. The method (300) as claimed in claim 3, wherein the extraction of the equidistant voltage values comprises:
identifying and sampling, by the one or more processor (204), the voltage measurements at regular and uniform intervals to create a dataset of the extracted equidistant voltage values.

6. The method (300) as claimed in claim 3, wherein the derivative of the extracted voltage comprises:
computing, by the one or more processor (204), the rate of change of the extracted voltage over time at the one or more of the selected equidistant points to capture the dynamic behaviour of the extracted voltage during the charging process.

7. The method (300) as claimed in claim 3, wherein the machine-learning model comprising, any or a combination of, a neural network, and predictive model trained on a large dataset of charging cycles from the lithium-ion cell.

8. The method (300) as claimed in claim 3, wherein the voltage drift configured to a gradual change in voltage characteristics of the cell during repeated charging cycles.

9. The method (300) as claimed in claim 3, wherein the charging process configured to monitor in real-time, and the charging data is continuously fed to the machine-learning model for dynamic discharge capacity estimation.

Documents