WO2023052910A1

WO2023052910A1 - System and method for estimating state of health and remaining useful life of a battery

Info

Publication number: WO2023052910A1
Application number: PCT/IB2022/058910
Authority: WO
Inventors: Ravish Kumar; Kunal KULKARNI; Rahul Jain
Original assignee: Kpit Technologies Limited
Priority date: 2021-10-01
Filing date: 2022-09-21
Publication date: 2023-04-06

Abstract

The present invention relates to a hybrid system and method forestimating the state of health (SOH) and remaining useful life (RUL) of Li-ion a battery of an electric vehicle. The system comprises sensors connected to the battery to monitor electrical attributes comprising current, voltage, temperature, and CC-CV switch count of the battery. A first deep learning computing unit (DLCU) that receives the values of electrical attributes as input and correspondingly predicts the SOH of the battery. A second DLCU is operatively coupled to the sensors and the first DLCU, which receives the values of the monitored electrical attributes and the predicted SOH as input and correspondingly predicts the SOH of the battery. The first and second DLCU comprise an LSTM network configured with an attention mechanism. The system removes deviating and inconsistent inputs and considers only consistent values of electrical attributes for SOH and RUL prediction, thereby making the system accurate and reliable.

Description

SYSTEM AND METHOD FOR ESTIMATING STATE OF HEALTH AND REMAINING USEFUL LIFE OF A BATTERY

TECHNICAL FIELD

[0001] The present invention relates to the field of battery management systems, and in particular, relates to a system and method for estimating the state of health (SOH) and remaining useful life (RUL) of a battery.

BACKGROUND

[0002] Rechargeable batteries have been widely employed in consumer devices, electric vehicles, and space technology because of their high energy/power density, low selfdischarge rate, and extended life. The automotive industry is currently experiencing a paradigm shift from conventional, diesel, and gasoline-propelled vehicles into second- generation hybrid and electric vehicles, with government and world organizations orienting and moving towards electric vehicles. Rechargeable batteries such as Lithium-ion batteries have been in demand as they are the most critical component in all electric vehicles.

[0003] Due to the inherent chemical reactions of the battery, the energy storage capacity of the battery may decrease as the battery is used, and the battery's performance generally deteriorates with each cycle of charging and discharging, resulting in sudden failure after its life cycle. In the case of sudden failure, an aged battery may result in substantial casualties and financial losses, especially in moving vehicles, military communication systems, navigation, aerospace, and other high-stakes situations. Consequently, the prognostics and health management (PHM) of a lithium-ion battery's state of health (SOH) and remaining useful life (RUL) are significant in lowering associated risks and costs of maintenance, thereby increasing the equipment reliability.

[0004] The SOH reflects the general condition of a battery and its ability to deliver the specified performance compared with a fresh battery. It takes into account such factors as charge acceptance, internal resistance, voltage, and self-discharge. SOH is a measure of the long-term capability of the battery and gives an indication not an absolute measurement, of how much of the available lifetime energy through put of the battery has been consumed, and how much is left. The battery’s RUL is defined as the remaining number of charge-discharge cycles of the battery with a specific output capacity until the battery reaches a certain degradation point, which helps estimate the degree of aging of the battery. [0005] Existing solutions available in the art predict the SOH of the battery using machine learning or deep learning-based methods, and then they use either linear interpolation or some mathematical modeling techniques for the RUL estimation. However, the existing solutions are inaccurate and inefficient as they fail to consider important input features of the battery, and considers irrelevant and deviating parameters as input features for SOH and RUL prediction.

[0006] One of the existing solutions uses an ensemble framework based on a convolutional bidirectional LSTM approach for SOH and RUL estimation, which considers time as an input feature in addition to other variables for estimating the SOH of the battery. However, when the battery is neither being charged nor discharged, using time as a feature will not account for the standing time of the battery, which is known to have an impact on the SOH. Besides, this solution only considers one of the parameters as an input at a time for SOH estimation, which makes the SOH estimation inaccurate.

[0007] The existing LSTM based approach then implements a forecasting method involving either linear interpolation or mathematical modeling technique, which uses the predicted SOH as the only input feature for RUL estimation. In addition, as the existing approach also uses discharge data or discharge trends of the battery as process features, this makes the existing approach inaccurate and inefficient as discharge trends of the battery drasticallyvary with the driving pattern of the vehicle and environmental conditions. Also, as the chargers for the battery may be different and may have different characteristics, as a result, the discharging patterns and input features monitored and considered for SOH and RUL estimation will be inconsistent throughout the process. Further, the existing solutions depend on data corresponding to the complete charging cycle of charging (which is 0-100% charging of the battery) and fail to check SOH during fractional charging cycle where the battery is charged from any non-zero charging level up to another non-zero charging level. Thus, the use of irrelevant, deviating, and insufficient data and parameters as input features in existing solutions make them inefficient and inaccurate in estimating SOH and RUL of the battery.

[0008] In addition, the existing solutions fail to consider the constant current to constant voltage (CC-CV) switch count while charging the battery in a certain threshold, in addition to the other input features, which is one of the most important parametersto be considered while predicting SOH and RUL of any rechargeable battery as shown in FIG. 5 and cannot be neglected. [0009] Thus, there is a need to overcome the above limitations, drawbacks, shortcomings, and provide an efficient, accurate, and reliable system and method for estimation of SOH and RUL of a battery used in electric vehicles, and the likes.

OBJECTS OF THE INVENTION

[0010] A general object of the present invention is to predict the SOH and RUL of a Lithium-Ion battery.

[0011] Another object of the present invention is to provide a system and method for estimation of SOH and RUL of a battery, which considers all the relevant input features of the battery such as current, voltage, temperature, and CC-CV count for accurate and efficient RUL estimation, which is computationally very simple as well as efficient and accurate, and which can be easily fine-tuned on the newer dataset.

[0012] Another object of the present invention is to count constant current to constant voltage (CC-CV) switch count while charging a battery.

[0013] Another object of the present invention is to provide a system and method for estimation of SOH and RUL of a battery, which is capable of checking SOH of the battery even during the fractional charging cycle, and which uses consistent pattern common to all types of chargers of the battery, giving accurate predictions on practical data.

SUMMARY

[0014] The present invention relates to the field of battery management systems, and in particular, relates to a system and method for estimating the state of health (SOH) and remaining useful life (RUL) of a battery.

[0015] According to an aspect, the system and method of the present disclosure comprise an input unit operatively coupled to a battery, afirst deep learning computing unit in communication with the input unit, and a second deep learning computing unit in communication with the input unit and the first deep learning computing unit. The first deep learning computing unit and the second deep learning computing unit have the same configuration. The input unit monitors the electrical attributes of the battery and generates an output based on the values ofthe electrical attributes comprising voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count of the battery, which is then provided to the first deep learning computing unit that predicts the SOH/ capacity of the battery based on the electrical attributes of the battery. The predicted SOH as well as the monitored electrical attributes of the battery are then fed to the second deep learning computing unit, which correspondingly processes the predicted SOH and electrical attributes to predict the RUL of the battery. The introduction of the second deep learning computing unitgreatly improves the accuracy of the RUL estimation by the proposed system and method, and also makes the fine-tuning of the corresponding computing units very easy.

[0016] Each of the first deep learning computing unit, and the second deep learning computing unit comprises an attentive long short-term memory (LSTM) sequential network having multiple nodes, an attention layer, and a dense layer. The attention layers assign weightage to different learning parameters and transfer the weighted learning parameters to the dense layers associated with the corresponding deep learning computing unit. The dense layer associated with the first deep learning computing unit predicts the SOH of the battery, and the dense layer associated with the second deep learning computing unit correspondingly predicts the RUL of the battery

[0017] In an aspect, prior to the transfer of the output data (or input features) of the input unit to the first deep learning computing unit, and the output of the first deep learning computing unit to the second deep learning computing unit for further processing and predicting the SOH and RUL, respectively, the input features are pre-processed to remove any inconsistent data and the consistent data is restructured into windowed form.

[0018] In an aspect, the CC-CV switch count of the battery is calculated based on a number of switching between a CC mode and a CV mode during a charging cycle of the battery when the voltage of the battery during the CC mode changes from a first predefined value (minimum value) to a second predefined value (maximum value), and current of the battery during the CV mode changes from a third predefined value (minimum value) to a fourth predefined value (maximum value). The method of calculating the CC-CV count monitoring the charging cycle of the battery and calculating a number of times when a minimum value of CC voltage and a maximum value of CV current is found to be in each charging cycle of the battery.

[0019] Accordingly, the proposed system and method utilize all the relevant and consistent input features comprising voltage, current, temperature, and CC-CV count for SOH estimation, and using the predicted SOH along with the other input features for RUL estimations, without relying on irrelevant and inconsistent features, thereby offering advantages of improved accuracy, faster and efficient computability, and robustness for RUL estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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.

[0021] FIG. 1 illustrates an exemplary block diagram of the proposed system for estimating the SOH and RUL of a battery according to an embodiment of the present invention.

[0022] FIG. 2A illustrates an exemplary architecture of the proposed system according to an embodiment of the present invention.

[0023] FIGs. 2B and 2C illustrates an exemplary architecture of the first deep learning computing unit, and the second deep learning computing unit of the system of FIG. 2A.

[0024] FIG. 2D illustrates an exemplary block diagram of LSTM network of the first deep learning computing unit, and the second deep learning computing unit according to an embodiment of the present invention.

[0025] FIG. 3 illustrates an exemplary flow diagram of the proposed method for estimating the SOH and RUL of a battery according to an embodiment of the present invention

[0026] FIG. 4A illustrates the CC and CV portions of a charging cycle of a battery for N number of samples.

[0027] FIG. 4B illustrates separate CC and CV portions of the charging cycle of FIG. 4 A for 10 different cycles.

[0028] FIG. 4C illustrates CC-CV switch point calculation using the FIGs. 4A and 4B.

[0029] FIG. 5 illustrates an exemplary graph depicting the factors affecting the SOH and RUL of a battery.

[0030] FIGs. 6A and 6B illustrate an exemplary plot depicting the predicted and actual SOH/Capacity of a battery, and the predicted and actual RUL of the battery, respectively, along with the corresponding evaluated values of MAE, RMSE, and R2 for a first sample of the dataset to test the SOH and RUL prediction capability of the proposed system and method.

[0031] FIGs. 7A and 7B illustrate an exemplary plot depicting the predicted and actual SOH/Capacity of a battery, and the predicted and actual RUL of the battery, respectively, along with the corresponding evaluated values of MAE, RMSE, and R2 for a second sample of the dataset to test the SOH and RUL prediction capability of the proposed system and method.

[0032] FIGs. 8 A and 8B illustrate an exemplary plot depicting the predicted and actual SOH/Capacity of a battery, and the predicted and actual RUL of the battery, respectively, along with the corresponding evaluated values of MAE, RMSE, and R2 for a third sample of the dataset to test the SOH and RUL prediction capability of the proposed system and method.

[0033] FIGs. 9A and 9B illustrate exemplary plots depicting the predicted andactual SOH/Capacity of the battery for fractional charging dataset to test the SOH prediction capability of the proposed system and method.

DETAILED DESCRIPTION

[0034] Embodiments of the present invention relate to a system and method for estimating the state of health (SOH) and remaining useful life (RUL) of a battery.

[0035] According to an aspect, the present disclosure elaborates upon a system for estimation of state of health (SOH) and remaining useful life (RUL) of a battery. The system comprises an input unit operatively coupled to the battery. The input unit is configured to monitor electrical attributes associated with the battery and correspondingly generate a first set of data packets. The electrical attributes comprise voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count. The system further comprises a first deep learning computing unit in communication with the input unit. The first deep learning computing unit comprising a first processor operatively coupled to a first memory comprising a first set of instructions executable by the first processor. The first deep learning computing unit is configured toreceive the first set of data packetsand extract the electrical attributes from the received first set of data packets, and predict a SOH of the battery based on the extracted electrical attributes, and correspondingly generate a second set of data packets indicative of the predicted SOH of the battery. The system further comprises a second deep learning computing unit in communication with the input unit and the first deep learning computing unit. The second deep learning computing unit comprising a second processor operatively coupled to a second memory comprising a second set of instructions executable by the second processor. The second deep learning computing unitis configured toreceive the first set of data packets from the input unit and extract the electrical attributes from the received first set of data packets, receive the second set of data packets from the first deep learning computing unit, and extract the predicted SOH of the battery, and predict the RUL of the battery based on the extracted electrical attributes and the predicted SOH and correspondingly generate a third set of data packets indicative of the predicted RUL of the battery.

[0036] In an embodiment, the system comprises a feature engineering unit configured to calculate the CC-CV switch count of the battery based on a number of switching between a CC mode and a CV mode during a charging cycle of the battery when the voltage of the battery during the CC mode changes from a first predefined value to a second predefined value, and current of the battery during the CV mode changes from a third predefined value to a fourth predefined value.

[0037] In an embodiment, each of the first deep learning computing units, and the second deep learning computing unit comprises a first long short-term memory (LSTM) sequential network, a second LSTM sequential network, an attention mechanism, and a dense layer. The attention mechanism assigns weightage to different learning parameters and transfers the weighted learning parameters to the dense layer associated with the first deep learning computing unit, and the second deep learning computing unit. The dense layer associated with the first deep learning computing unit correspondingly predicts the SOH of the battery, and the dense layer associated with the second deep learning computing unit correspondingly predicts the RUL of the battery.

[0038] In an embodiment, the system comprises a pre-processing unit operatively coupled to the input unit and the first deep learning computing unit. The pre-processing unit is configured to monitor values of the monitored electrical attributes for a predefined number of samples, and remove inconsistent values of the electrical attributes from the predefined number of samples to generate the first set of data packets comprising consistent values of the electrical attributes. Further, the system comprises a restructuring unit operatively coupled to the input unit, the first deep learning computing unit, and the second deep learning computing unit. The restructuring unit is configured to restructure the first set of data packets, and the second set of data packets into a windowed form, and correspondingly transfer the windowed form of the first set of data packets, and the second set of data packets to the first deep learning computing unit, and the second deep learning computing unit, respectively.

[0039] In an embodiment, the input unit comprises a set of sensors operatively coupled to the battery and comprising any or a combination of a current sensor, voltage sensor, and temperature sensor.

[0040] In an embodiment, the battery is a Li-ion battery associated with an electric vehicle. [0041] According to another aspect, the present disclosure elaborates upon a system for estimation of state of health (SOH) and remaining useful life (RUL) of a battery-operated electric vehicle (BEV). The system comprises an input unit operatively coupled to a battery of the BEV. The input unit is configured to monitor electrical attributes associated with the battery and correspondingly generate a first set of data packets. The electrical attributes comprise voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count. The system further comprises a first deep learning computing unit in communication with the input unit. The first deep learning computing unit comprising a first processor operatively coupled to a first memory comprising a first set of instructions executable by the first processor. The first deep learning computing unit is configured to receive the first set of data packets and extract the electrical attributes from the received first set of data packets, and predict a SOH of the battery based on the extracted electrical attributes, and correspondingly generate a second set of data packets indicative of the predicted SOH of the BEV. The system further comprises a second deep learning computing unit in communication with the input unit and the first deep learning computing unit. The second deep learning computing unit comprising a second processor operatively coupled to a second memory comprising a second set of instructions executable by the second processor. The second deep learning computing unit is configured to receive the first set of data packets from the input unit and extract the electrical attributes from the received first set of data packets, receive the second set of data packets from the first deep learning computing unit, and extract the predicted SOH of the battery, and predict the RUL of the battery based on the extracted electrical attributes and the predicted SOH and correspondingly generate a third set of data packets indicative of the predicted RUL of the BEV.

[0042] According to another aspect, the present disclosure elaborates upon a method for estimation of state of health (SOH) and remaining useful life (RUL) of a battery. The method comprises a step of receiving, by a first deep learning computing unit, a first set of data packets pertaining to electrical attributes associated with the battery, where the electrical attributes comprise voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count. The method comprises a step of predicting, by the first deep learning computing unit, a SOH of the battery based on the received first set of data packets, and correspondingly generating a second set of data packets. Further, the method comprises a step of receiving, by a second deep learning computing unit in communication with the input unit and the first deep learning computing unit, the first set of data packets and the second set of data packets, and correspondingly extracting the predicted SOH of the battery and the electrical attributes of the battery, followed by a step of processing, by the second deep learning computing unit, the extracted electrical attributes and the predicted SOH of the battery to predict a RUL of the battery.

[0043] In an embodiment, the method of calculating the CC-CV count comprises the step of monitoring the charging cycle of the battery and calculating a number of times when a minimum value of CC voltage and a maximum value of CV current are found to be in each charging cycle of the battery.

[0044] In an embodiment, the method comprises the steps of monitoring, by an input unit operatively coupled to the battery, the electrical attributes of the battery, and monitoring, by a pre-processing unit, values of the monitored electrical attributes for a predefined number of samples, and removing inconsistent values of the electrical attributes from the predefined number of samples to generate the first set of data packets comprising consistent values of the electrical attributes.

[0045] Referring to FIG. 1, in an aspect, the present disclosure elaborates upon a system 100 for estimating the SOH and RUL of a battery 102. The system 100 comprises an input unit, a pre-processing unit 104, a feature engineering unit 106, a first restructuring unit 108-1, a second restructuring unit 108-2, a first deep learning computing unit 110, and a second deep learning computing 112. The battery 102 herein refers to a single battery/cell, or a pack of batteries, but not limited to the like. In an exemplary embodiment, the battery 102 is a Lithium-ion battery associated with a battery-operated electric vehicle (BEV), but not limited to the like. The input unit comprises a set of sensors selected from a current sensor, voltage sensor, and temperature sensor, which is operatively coupled to the battery and configured to monitor electrical attributes comprising voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count associated with the battery. The CC-CV switch count pertains to anumber of times when a minimum value of CC voltage and a maximum value of CV current is found to be in each charging cycle of the battery. The preprocessing unit 104 is configured with the input unit, which monitors the values of the monitored electrical attributes of the battery 102 for a predefined number of samples and removesany inconsistent values of the electrical attributes from the predefined number of samples to generate a first set of data packets comprising consistent values of the electrical attributes.

[0046] In an embodiment, the pre-processing unit 104 examines the inconsistencies by examining the curve plots of each electrical attribute for consecutive samples. If a spike in the curve plot of the electrical attribute is detected, the values of the spiked electrical attribute is discarded from further processing by the system 100. For example, if the voltage for 10 consecutive samples is around 2V and suddenly at 11th sample, the voltage value becomes 4 V and again at 12th sample value comes to be 2V, then voltage value at 11th sample is not consistent, and hence must be removed. In another embodiment, if the magnitude of any electrical attribute goes beyond a defined range, the values of the out-of-range electrical attribute are discarded from further processing by the system 100. For example, if discharging of the battery is happening at a fixed current value of -2 A, and there are some values of current having -2.5 A magnitude, then such samples are removed by the preprocessing unit 104.

[0047] In an embodiment, the feature engineering unit 106 is configured with the preprocessing unit 104, which is operable to calculate the CC-CV switch count of the battery based on a number of switching between a CC mode and a CV mode during a charging cycle of the battery when the voltage of the battery during the CC mode changes from a first predefined value (minimum value) to a second predefined value (maximum value), and current of the battery during the CV mode changes from a third predefined value (minimum value) to a fourth predefined value (maximum value).

[0048] In an embodiment, the first restructuring unit 108-1 receives the first set of data packets as input features comprising the consistent values of electrical attributes of the battery 102 and restructures the received first set of data packets into a windowed form, and correspondingly transfers the windowed form of the first set of data packets to the first deep learning computing unit 110, where the first deep learning computing unit 110 processes the received consistent values of electrical attributes to predict the SOH of the battery 102, and correspondingly generate a second set of data packets. For instance, if SOH and RUL are to be predicted for a 1^st cycle of charging, then the first deep learning computing unit 110 takes the four input features (i.e current, voltage, temperature, and CC-CV switch count of battery) in a window of a certain size from the first restructuring unit 108-1 for the 1^st cycle, and predict the SOH of the battery for the particular cycle.

[0049] In an embodiment, the second restructuring unit 108-2 also receives the first set of data packets comprising the consistent values of electrical attributes from the feature engineering unit 106 and further receives the second set of data packets corresponding to the predicted SOH of the battery from the first deep learning computing unit 110. The second restructuring unit 108-2 restructures the received first set of data packets, and the second set of data packets into a windowed form, and correspondingly transfers the windowed form of the first and second set of data packets to the second deep learning computing unit 112, where the second deep learning computing unit 112 processes the received consistent values of electrical attributes and the predicted SOH to predict the RUL of the battery 102. Here, the second deep learning computing unit 112 takes the five input features (i.e current, voltage, temperature, CC-CV switch count of battery, and the predicted SOH) in a window of a certain size from the second restructuring unit 108-2 and correspondingly predict the RUL of the battery 102.

[0050] In an exemplary embodiment, the first restructuring unit 108-1 is the same as the second restructuring unit 108-2. In another exemplary embodiment, the first restructuring unit 108-1 and the second restructuring unit 108-2 are different.

[0051] Referring to FIG. 2A, the architecture of the proposed system 100 is disclosed. System 100 comprises one or more processor(s) 202 operatively coupled to a memory 204 storing computer-readable instructions. The one or more processor(s) 202 are implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuitries, and/or any devices that manipulate data based on operational instructions. Among other capabilities, one or more processor(s) 202 are configured to fetch and execute computer-readable instructions stored in the memory 204. Memory 204 comprises any non-transitory storage device comprising, for example, volatile memory such as RAM, or non-volatile memory such as EPROM, flash memory, and the like. In an embodiment, systemlOO also comprises an interface(s) 206. The interface(s) 206 comprises a variety of interfaces, for example, interfaces for data input and output devices referred to as I/O devices, storage devices, and the like. In an embodiment, system lOOcomprises a communication unit operatively coupled to one or more processor(s) 202. The communication unit is configured to communicatively couple the system 100 to the battery, and other devices, as well as enable communication between components of the system 100. In an exemplary embodiment, the communication unit comprises any or a combination of Bluetooth module, NFS Module, WIFI module, transceiver, and wired media, but not limited to the likes. In an embodiment, the processing engine(s) 208 are implemented as a combination of hardware and programming (for example, programmable instructions) to implement one or more functionalities of the processing engine(s) 208. Database 210 comprises data that is either stored or generated as a result of functionalities implemented by any of the components of the processing engine(s). In an embodiment, the processing engine(s) 208 comprises a battery attribute extraction unit 212, a pre-processing unit 214, a feature engineering unit 216, a restructuring unit 218, a first deep learning computing unit 220, a second deep learning computing unit 222, and other units (s), but not limited to the likes. The other unit(s) implements functionalities that supplement applications or functions performed by the system 100 or the processing engine(s) 208. The data (or database 210) serves, amongst other things, as a repository for storing data processed, received, and generated by one or more of the units.

[0052] In an embodiment, the battery attribute extraction unit 212 enables the processors to actuate the input unit to monitor electrical attributes comprising voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count associated with the battery, and correspondingly receive input features pertaining to the monitored electrical attributes.

[0053] In an embodiment, the pre-processing unit 214 enables the processors 202 toreceive and monitor the values of the electrical attributes of the battery 102 for a predefined number of samples, and remove any inconsistent values of the electrical attributes from the predefined number of samples and generate a first set of data packets comprising consistent values of the electrical attributes. The pre-processing unit 214 examines the inconsistencies by examining the curve plots of each electrical attribute for consecutive samples. If a spike in the curve plot of the electrical attribute is detected, the values of the spiked electrical attribute is discarded from further processing by the system. Further, if the magnitude of any electrical attribute goes beyond a defined range, the values of the out-of-range electrical attribute are discarded from further processing by the system 100.

[0054] In an embodiment, the feature engineering unit 216 enables the processors 202 to calculate the CC-CV switch count during the charging cycle of the battery. Referring to FIGs. 4A to 4C, in an exemplary implementation, the current and voltage plot of the battery for 10 cycles as shown in FIG. 4A was examined to determine and check the CC-CV switch count by the proposed system. Further, to find whether the switching between constant current and constant voltage is happening at single point, both the current and voltage plots were brought on the same scale and were separated and checked for 10 different cycles as shown in FIG. 4B. During, CC-CV charging of any battery, the battery 102 starts charging with the constant current (CC) mode until the voltage reaches a certain value, usually 4.2 V. The charging then switches to the constant-voltage (CV) mode until the current reaches a certain value, usually 0A.

[0055] The processors 202 calculate the CC-CV switch count of the battery 102 based on a number of switching between a CC mode and a CV mode during a charging cycle of the battery when the voltage of the battery during the CC mode changes from a first predefined value (minimum value) to a second predefined value (maximum value), and current of the battery during the CV mode changes from a third predefined value (minimum value) to a fourth predefined value (maximum value). The conditions showing CC-CV switching for the battery are stated in Table- 1 below for three exemplary datasets.

TABLE-1

number of CC-CV switches when the voltage of the battery in the CC portion goes from 4 V to 4.2 V, and current in the CV portion goes from -1A to -2A.

[0056] In an embodiment, the restructuring unit 218 enables the processors 202 to receive the first set of data packets as input (input features) comprising the consistent values of electrical attributes of the battery 102 and further enables the processors 202 to restructure the received first set of data packets into a windowed form.

Before restructuring, the input x at any time step t is defined by equation 1 below, x_t = [I_t, V_t, T_t, CC-CVJ

After restructuring, the restructured input features at time t is defined by

Xl_t = [xt-_w+i, x_t-w, > , x_t-i, x_t], where x_t is same as the equation 1. W is the window size and be tuned as per the dataset.

[0057] In an embodiment, the first deep learning computing unit 220 enables a first processor among the processors 202 to process the windowed form of the restructured input features comprising the electrical attributes (current, voltage, temperature, and CC-CV switch count) to predict the SOH of the battery for a particular charging cycle, and correspondingly generate a second set of data packets indicative of the predicted SOH of the battery 102.

[0058] In an embodiment, the restructuring unit 218 enables the processors 202 to receive the first set of data packets and the second set of data packets as input features comprising the consistent values of electrical attributes and the predicted SOH and further enables the processors 202 to restructure the received first and second set of data packets into a windowed form.

Before restructuring, the input x at any time step t is defined by equation 2 below, x_t = [It, V_t, T_t, CC-CVt, Predicted_SOH_t]

After restructuring, the restructure input features at time t is defined by

X2_t = [xt-_w+i, x_t-w, > , x_t-i, x_t], where x_t is same as the equation 2. W is the window size and be tuned as per the dataset. [0059] In an embodiment, the second deep learning computing unit 222 enables a second processor among the processors 202 to process the windowed form of the restructured input features comprising the electrical attributes (current, voltage, temperature, and CC-CV switch count) and the predicted SOH to predict the RUL of the battery, and correspondingly generate a third set of data packets indicative of the predicted RUL of the battery 102.

[0060] FIG. 2B illustrates an exemplary architecture of the first deep learning computing unit 220 of the proposed system.FIG. 2C illustrates an exemplary architecture of the second deep learning computing unit 222 of the proposed system.

[0061] As illustrated, in an embodiment, the architecture of the first deep learning computing unit 220 and the second deep learning computing unit 22 are the same.The first deep learning unit 220 comprises a first Long Short-Term Memory (LSTM) sequential network unit220-l having multiple nodes (100 nodes, but not limited to the like), and a second LSTM sequential network unit 220-2 having multiple nodes (50 nodes, but not limited to the like). Further, the second deep learning unit 222 also comprises a first LSTM sequential network unit 222-1 having multiple nodes (100 nodes, but not limited to the like), and a second LSTM sequential network unit 222-2 having multiple nodes (50 nodes, but not limited to the like).

[0062] An exemplary architecture of a general LSTM node is shown in FIG. 2D. The two LSTMs of each deep learning computing unit 220. 222 are stacked, which helps the deep learning computing units 220, 222 to extract deeper insights from the received data packets and learn the pattern to accurately predict the SOH and RUL of the battery 102.

[0063] In an embodiment, each of the deep learning units 220, 222 comprise an attention layer 220-3, 222-3 (also referred to as attention mechanism, herein) that ensures the accumulation of part of learning needed to attend over while reading the learning done by two stacked LSTMs (LSTM1 and LSTM2). In addition, the attention layer 220-3, 222-3 helps the dense layer 220-4, 222-4 to capture the whole traffic dynamics (like differentiation, crosscorrelation) among the different dimensions of learning. The attention layer220-3, 222-3 assigns the weightage to different learning parameters and sends it to the dense layer 220-4, 222-4.

[0064] In an embodiment, the dense layer 220-4, 222-4 is a fully connected layer that observes all inputs provided to it and mapsthe inputs to the corresponding output. Suppose at the dense layer, 100 x 100 matrix enters as an input, but there is only 1 capacity value is output. The dense layer then maps the 100 x 100 matrix to a single capacity. The dense layer receives the output of the attention layer as input along with weightage assigned to it and generates the final output in the desired format whether it is the SOH or RUL of the battery.

[0065] The attention layer 220-3, 222-3 assigns weightage to different learning parameters and transfers the weighted learning parameters to the dense layers 220-4, 222-4 associated with the first deep learning computing unit 220, and the second deep learning computing unit 222. Further, the dense layer 220-4 associated with the first deep learning computing unit 220 correspondingly predicts the SOH of the battery 102, and the dense layer 222-4 associated with the second deep learning computing unit 222 correspondingly predicts the RUL of the battery 102.

[0066] Referring to FIG. 3, in another aspect, the present disclosure elaborates upon a method 300 for estimating the SOH and RUL of a battery. Method 300 involves an input unit comprising sensors being operatively coupled to the battery, which is in communication with a first deep computing unit for accurately estimating the SOH of the battery. Further, the method involves a second deep computing unit, which is in communication with the first deep computing unit and the input unit.

[0067] In an embodiment, method 300 comprises the step of monitoring, by the input unit, electrical attributes comprising voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count associated with the battery. Further, method 300 comprises the step of monitoring, by a pre-processing unit, values of the monitored electrical attributes for a predefined number of samples, and removing inconsistent values of the electrical attributes from the predefined number of samples to generate the first set of data packets comprising consistent values of the electrical attributes.

[0068] In an embodiment, method 300 comprises step 302 of receiving, by the first deep learning computing unit, the first set of data packets pertaining to the electrical attributes from the pre-processing unit.Further, method 300 comprises step 304 of predicting, by the first deep learning computing unit, a SOH of the battery based on the received first set of data packets, and correspondingly generating a second set of data packets.

[0069] Method 300 further comprises step 306 of receiving, by the second deep learning computing unit that is in communication with the input unit and the first deep learning computing unit, the first set of data packets and the second set of data packets, and correspondingly extracting the predicted SOH of the battery and the electrical attributes of the battery. Further, method 300 comprises step 308 of processing, by the second deep learning computing unit, the extracted electrical attributes, and the predicted SOH of the battery to predict the RUL of the battery. TEST RESULTS AND VALIDATION

[0070] The SOH and RUL prediction capability and accuracy of the proposed system 100 and method 300 was validated by implementing the proposed system 100 and method 300 in conditions stated in Table 1 for Sample 1-3 datasets to predict the corresponding SOH and RUL and comparing the predicted SOH and RUL with the actual test results.

[0071] As SOH estimation is a regression problem, to estimate the performace of regression models, known methods such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and the R-Squared (R )value are used.

[0072] Mean Absolute Error (MAE) is a difference between predicted value of SOH and the actual value of SOH over n samples. Lesser the MAE, better will be the performance. MAE (Mean Absolute Error) = (1/n)* Xi"(SoH_prcdictcd(i) - SoH_actual(i)).

[0073] Root Mean Squared Error (RMSE) is square of difference between predicted value of SOH and the actual value of SOH over n samples. Lesser the RMSE, better will be the model. RMSE (RootMean Square Error) = [(1/n)* Xi"(SoH_prcdictcd(i) - SoH_actual(i))²]⁰'⁵.

[0074] R-squared (R2) is defined in terms of the explainability of variance of one variable with respect to the variance of other variable. R (R-Squared) = (1 - (Model Mean Squared Error )/(Baseline Mean Squared Error)), Where Model MSE is mean of square errors between predicted SOH and actual SOH, and Baseline MSE is the mean of square errors between actual SOH and mean of actual SOH. The values of R lies in the range of (-co, 1], More the value closer to 1, more better the model.

[0075] FIGs. 6A and 6B illustrate an exemplary plot depicting the predicted and actual SOH/Capacity of a battery, and the predicted and actual RUL of the battery, respectively, along with the corresponding evaluated values of MAE, RMSE, and R2 for sample 1 of the dataset to test the SOH and RUL prediction capability of the proposed system 100 and method 300. Table-2 below shows the comparison between the predicted RUL and actual RUL of the battery after different cycles for sample 1.

TABLE-2

[0076] FIGs. 7A and 7B illustrate an exemplary plot depicting the predicted and actual SOH/Capacity of a battery, and the predicted and actual RUL of the battery, respectively, along with the corresponding evaluated values of MAE, RMSE, and R2 for sample 2 of the dataset to test the SOH and RUL prediction capability of the proposed system 100 and method 300. Table-3 below shows the comparison between the predicted RUL and actual RUL of the battery after different cycles for sample 2. TABLE-3

[0077] FIGs. 8 A and 8B illustrate an exemplary plot depicting the predicted and actual SOH/Capacity of a battery, and the predicted and actual RUL of the battery, respectively, along with the corresponding evaluated values of MAE, RMSE, and R2 for sample 3 of the dataset to test the SOH and RUL prediction capability of the proposed system 100 and method 300. Table-4 below shows the comparison between the predicted RUL and actual

RUL of the battery after different cycles for sample 3.

TABLE-4

[0078] To check the SOH accuracy of the proposed system 100 and method 300 during fractional cycling when the battery is not charged from 0 to 100% SOC and starts from any non-zero start percentage to an end percentage, a modified dataset was constructed using the

3 batteries from sample 1 dataset considering one cycle of the battery, assuming 200 samples.

The four possibilities for real life uage of the battery is given in Table 5 below.

TABLE-5

[0079] The cycles are selected at random to fulfill any one of the above four conditions of Table 5. For example, cycle 1 may have 0-60% sample. Cycle 2 may have 25-75% samples, and so on for other cycles. When the driver code for modifications on three batteries were run, different battery sets were found every time. The 9 battery sets were made from b5, b6 and b7, represented by b5[0], b5[l], b5[2] ; b6[0], b6[l], b6[2]; b7[0], b7[l], b7[2], [0080] All the 9 battery sets were unique and had random numbers of samples per cycle. Hence, all the battery sets have different capacity at different times, as well as unique CC-CV switch values.

Training set=b5[0], b5[l], b5[2]; b6[0], b6[l], b6[2]

Testing set=b7[0], b7[1], b7[2]

[0081] .FIGs. 9A illustrates exemplary plot depicting the predicted and actual SOH/Capacity of the battery for fractional cycling dataset b7[l], along with the corresponding evaluated values of MAE, RMSE, and R2. FIGs. 9B illustrates exemplary plot depicting the predicted and actual SOH/Capacity of the battery for original dataset b7 to test the SOH prediction capability and flexibility of the proposed system 100 and method 300 [0082] Accordingly, system 100 and method 300 of the present invention utilize all the relevant and consistent input features comprising voltage, current, temperature, and CC-CV count for SOH estimation, and using the predicted SOH along with the other input features for RUL estimations, without relying on irrelevant and inconsistent features, thereby offering advantages of improved accuracy, faster and efficient computability, and robustness for RUL estimation.

ADVANTAGES OF THE PRESENT INVENTION

[0083] The present invention predicts the SOH and RUL of a Lithium-Ion battery.

[0084] The present invention provides a system and method for estimation of SOH and

RUL of a battery, which considers all the relevant input features of the battery such as current, voltage, temperature, and CC-CV count for accurate and efficient RUL estimation, which is computationally very simple as well as efficient and accurate, and which can be easily fine-tuned on the newer dataset.

[0085] The present invention counts constant current to constant voltage (CC-CV) switch count while charging a battery. [0086] The present invention provides a system and method for estimation of SOH and RUL of a battery, which is capable of checking SOH of the battery even during the fractional charging cycle, and which uses consistent patetrn common to all types of chargers of the battery, giving accurate predictions on practical data.

Claims

We Claim:

1. A system for estimation of state of health (SOH) and remaining useful life (RUL) of a battery, the system comprising: an input unit operatively coupled to the battery, the input unit configured to monitor electrical attributes associated with the battery and correspondingly generate a first set of data packets, wherein the electrical attributes comprise voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count; a first deep learning computing unit in communication with the input unit, the first deep learning computing unit comprising a first processor operatively coupled to a first memory comprising a first set of instructions executable by the first processor, and configured to: receive the first set of data packets, and extract the electrical attributes from the received first set of data packets; and predict a SOH of the battery based on the extracted electrical attributes, and correspondingly generate a second set of data packets indicative of the predicted SOH of the battery; and a second deep learning computing unit in communication with the input unit and the first deep learning computing unit, the second deep learning computing unit comprising a second processor operatively coupled to a second memory comprising a second set of instructions executable by the second processor, and configured to: receive the first set of data packets from the input unit, and extract the electrical attributes from the received first set of data packets; receive the second set of data packets from the first deep learning computing unit, and extract the predicted SOH of the battery; and predict the RUL of the battery based on the extracted electrical attributes and the predicted SOH and correspondingly generate a third set of data packets indicative of the predicted RUL of the battery.

2. The system as claimed in claim 1, wherein the system comprises a feature engineering unit configured to calculate the CC-CV switch count of the battery based on a number of switching between a CC mode and a CV mode during a charging cycle of the battery when the voltage of the battery during the CC mode changes from a first predefined value to a second predefined value, and current of the battery during the CV mode changes from a third predefined value to a fourth predefined value. The system as claimed in claim 1, wherein each of the first deep learning computing unit, and the second deep learning computing unit comprises a first long short-term memory (LSTM) sequential network, a second LSTM sequential network, an attention mechanism, and a dense layer; wherein the attention mechanism assigns weightage to different learning parameters and transfers the weighted learning parameters to the dense layer associated with the first deep learning computing unit, and the second deep learning computing unit, and wherein the dense layer associated with the first deep learning computing unit correspondingly predicts the SOH of the battery, and the dense layer associated with the second deep learning computing unit correspondingly predicts the RUL of the battery. The system as claimed in claim 1, wherein the system comprises: a pre-processing unit operatively coupled to the input unit and the first deep learning computing unit, the pre-processing unit configured to monitor values of the monitored electrical attributes for a predefined number of samples, and remove inconsistent values of the electrical attributes from the predefined number of samples to generate the first set of data packets comprising consistent values of the electrical attributes; and a restructuring unit operatively coupled to the input unit, the first deep learning computing unit, and the second deep learning computing unit, the restructuring unit is configured to restructure the first set of data packets, and the second set of data packets into a windowed form, and correspondingly transfer the windowed form of the first set of data packets, and the second set of data packets to the first deep learning computing unit, and the second deep learning computing unit, respectively. The system as claimed in claim 1, wherein the input unit comprises a set of sensors operatively coupled to the battery and comprising any or a combination of a current sensor, voltage sensor, and temperature sensor. The system as claimed in claim 1, wherein the battery is a Li-ion battery associated with an electric vehicle. A system for estimation of state of health (SOH) and remaining useful life (RUL) of a battery-operated electric vehicle (BEV), the system comprising: an input unit operatively coupled to a battery of the BEV, the input unit configured to monitor electrical attributes associated with the battery and correspondingly generate a first set of data packets, wherein the electrical attributes comprise voltage, current, temperature, and constant current to constant voltage (CC- CV) switch count; a first deep learning computing unit in communication with the input unit, the first deep learning computing unit comprising a first processor operatively coupled to a first memory comprising a first set of instructions executable by the first processor, and configured to: receive the first set of data packets, and extract the electrical attributes from the received first set of data packets; and predict a SOH of the battery based on the extracted electrical attributes, and correspondingly generate a second set of data packets indicative of the predicted SOH of the battery of the BEV; and a second deep learning computing unit in communication with the input unit and the first deep learning computing unit, the second deep learning computing unit comprising a second processor operatively coupled to a second memory comprising a second set of instructions executable by the second processor, and configured to: receive the first set of data packets from the input unit, and extract the electrical attributes from the received first set of data packets; receive the second set of data packets from the first deep learning computing unit, and extract the predicted SOH of the battery; and predict the RUL of the battery based on the extracted electrical attributes and the predicted SOH and correspondingly generate a third set of data packets indicative of the predicted RUL of the battery of the BEV. A method for estimation of state of health (SOH) and remaining useful life (RUL) of a battery, the method comprising the steps of: receiving, by a first deep learning computing unit, a first set of data packets pertaining to electrical attributes associated with the battery, the electrical attributes comprising voltage, current, temperature, and constant current to constant voltage (CC-CV) switch count; predicting, by the first deep learning computing unit, a SOH of the battery based on the received first set of data packets, and correspondingly generating a second set of data packets; receiving, by a second deep learning computing unit in communication with the input unit and the first deep learning computing unit, the first set of data packets and the second set of data packets, and correspondingly extracting the predicted SOH of the battery and the electrical attributes of the battery; and processing, by the second deep learning computing unit, the extracted electrical attributes and the predicted SOH of the battery to predict a RUL of the battery.

9. The method as claimed in claim 8, wherein the method of calculating the CC-CV count comprises the step of monitoring the charging cycle of the battery and calculating a number of times when a minimum value of CC voltage and a maximum value of CV current is found to be in each charging cycle of the battery. 10. The method as claimed in claim 8, wherein the method comprises the steps of: monitoring, by an input unit operatively coupled to the battery, the electrical attributes of the battery; and monitoring, by a pre-processing unit, values of the monitored electrical attributes for a predefined number of samples, and removing inconsistent values of the electrical attributes from the predefined number of samples to generate the first set of data packets comprising consistent values of the electrical attributes.