Title of Invention | A PROCESS OF DETERMINING STATE OF CHARGE AND STATE OF HEALTH OF A BATTERY |
---|---|
Abstract | A process of determining State of Charge and State of Health of a battery under on-load conditions comprises the steps of drawing the temperature-internal resistance characteristics of a battery by measuring the battery temperature and the internal resistance at three to four operating temperatures with different ages of battery at various levels of discharge conditions; training an artificial neural network with said measured inputs, namely, internal resistance and battery temperature, and with other inputs, duly measured, and selected from one or more of terminal voltage, current drawn from battery and consumption time, for the purpose of obtaining the output Specific Gravity of the battery; determining the State of Charge of the battery by deploying fuzzy rule base using the battery temperature, as measured, and the specific gravity; and determining the State of Health of the battery in terms of percentage of remaining life of the battery, based on the measured inputs such as, current drawn, consumption time, total consumption time and specific gravity, and using a mathematical model. |
Full Text | Field of the Invention The present invention relates generally to a method for determining the charge condition and remaining capacity of a battery, and more specifically, to an intelligent process for estimating the State of Charge and State of Health of a battery under online conditions using a Neuro-fuzzy model. Background of the Related Art Battery, an electrochemical device, is a complex system and is difficult to model with conventional methods using mathematical model, circuit model or any other means. Till now no study exists to measure or determine the State of Charge (SoC) of battery under onload conditions. The state- of-health (SoH) of battery provides the capability of a battery to meet its load demand for ages. Exemplary of prior art setups which have been proposed for determination of SoH are those described in US Patent Specifications 5,365,453; 4,080,560; 5,159,272; US 6,668,247B2 and 6,456,988. US 5,365,453 discloses a method in which ratio of change in battery voltage to a change in load is used to predict impending battery failure in battery powered electronic devices. Similar method in which the battery response to and recovery from the application of a load, is used to determine the SoH of batteries, is reported in US 4,080,560 and 5,159,272. Another approach has been described in US 6,668,247, wherein the SoH is determined from a fuzzy system trained in a relationship between the impedance characteristic of battery at at least one selected frequency, and the number of complete charge and discharge cycles that the battery undergoes and characteristic of internal parameters of the battery. Yet another approach has been taken in the US 6,456,988, wherein the determination of SoH of electrochemical device is described by using fuzzy logic trained in relationship between Internal characteristic parameters (at external operation and environmental conditions) and characteristic parameters of the load. The SoC and SoH have been determined in the prior art by employing Fuzzy logic with measurement of internal/external parameters. Each of the prior art methods for determining the SoC and SoH of a battery, as described above, suffers from one limitation or the other. Summary of the Invention The inherent drawbacks and deficiencies of the prior art methoos/systems are overcome or alleviated by the process according to the present invention, for determining the SoC and SoH of a battery. The object of the present invention is to concentrate on the modeling of nonlinear behavior of battery, not through mathematical/ algorithmic approach like the prior work in the field of the invention, but to simulate the entire process based on real data, using more significant battery parameters. Said 'Nonlinear behavior mapping' of battery has been conducted using Artificial Neural Network (ANN). This has been found to provide a more reliable and accurate estimation of SoC and SoH of a battery. The process according to the present invention has been demonstrated by using observations on 12 volt Lead-Acid battery used in cars. The SoC of battery has been determined by modeling the non-linear relationship of input parameters, viz., internal resistance,battery temperature, consumption time, terminal voltage and current drawn with specific gravity at various operating temperatures and charged state of battery in question of batteries of various ages. The specific gravity of a battery is the true representation of SoC of the battery. This relationship is then modeled using fuzzy system which linguistically represents the State of Charge of the battery as fully charged, more than half charged, half charged, less than half charged, flat charged, etc. The State of Health (SoH) of a battery provides the capability of a battery to meet its load demand for ages. To obtain the online SoH of a battery, a mathematical relationship is drawn between the online specific gravity, consumption time, total consumption time and current drawn giving, on percentage basis, the remaining life of the battery. Accordingly, the present invention provides a process of determining State of Charge and State of Health of a battery under on-load conditions, said process comprising the steps of : drawing the temperature-internal resistance (temp-IR) characteristics of the battery by measuring the battery temperature and the internal resistance at three to four operating temperatures with different ages of battery at various levels of discharge conditions; training an artificial neural network (ANN) with measured inputs, namely, terminal voltage, current drawn from the battery and internal resistance, and, optionally, with other inputs, duly measured, and selected from one or more of, battery temperature and consumption time, for the purpose of obtaining the output Specific Gravity of the battery, said battery temperature input to the ANN being extracted by least square fit of said (temp-IR) characteristics; determining the State of Charge of the battery by deploying fuzzy rule base using the battery temperature, as extracted, and the Specific Gravity, as found in the preceding step; and determining the State of Health of the battery in terms of percentage of remaining life of the battery, based on the measured inputs such as, current drawn from the battery, consumption time, total consumption time and slope of said Specific Gravity to total consumption time, and using a mathematical model, such as herein described. The present invention provides these and other advantages as will be apparent to those having ordinary skill in the art upon careful reading of the following detailed description of the invention with respect to the accompanying drawings described below. Brief Description of the Accompanying Drawings The invention is described with reference to the accompanying drawings, of which: Fig. 1 is a block diagram of the process model for determining the State of Charge and State of Health of a battery; Fig. 2 shows the fully connected architecture of the Multi layer Perceptron (MLP) of Artificial Neural network employed in modeling; Figure 3 shows a Sigmoid function used as an activation function at hidden layers of ANN; Figures 4A, 4B and 4C show the membership functions for the Fuzzy variables (A) Temperature, (B) Specific Gravity, and (C) State of Charge; Table-1 is a table of Rule Specification of Fuzzy indicators employed by the Fuzzy model for the determination of the State of Charge of a battery in accordance with the present invention; Table-2 shows weights and biases of synaptic connection of ANN for complete trained model for Lead-Acid battery Model MF40S; Table-3 shows the most accurate ANN training that can be achieved with at least five preferred parameters. Detailed Description of the Preferred Embodiments The present invention describes an intelligent process for determining the State of Charge (SoC) and State of Health (SoH) of any type of battery under online conditions using a Neuro-Fuzzy technique. Such a technique involves mapping of the complex and nonlinear behavior of a battery through training of Artificial Neural Network (ANN), which due to its in-built capacity, handles the non-linearity of the battery efficiently and facilitates modeling said nonlinear behavior of battery more closely so as to provide an accurate estimation of SoC and SoH. The indication of SoC and SoH of a Battery depends on large number of parameters, most of which are impossible to detect in onload condition or without the support of sufficient laboratory instruments. One such parameter is the Specific Gravity of a battery which acts as the true representation of SoC of the battery. Specific gravity of a battery, which is monitored through a chemical process, is also among the other parameters, which is difficult to be measured online. Therefore, other parameters of the battery are used, in accordance with the present invention, to provide an indirect indication of Specific Gravity, corrosion, etc. and in turn SoC and SoH, as described hereinafter. Fig. 1 shows a schematic diagram of the generalized process model for determining the SoC and SoH of a battery according to a preferred embodiment of the invention. The process uses a Neuro-fuzzy model wherein an Artificial Neural network (ANN) is; first trained to simulate the discharging process of a specific battery and then the output of the ANN is translated into indicating the State of Charge of battery in linguistic terminology using fuzzy rule base. ANN is adapted to produce at its output, the corresponding specific gravity of the battery. For this purpose, the ANN is adapted for a suitable representation of the behavior of the battery. In a preferred embodiment, the process has been demonstrated using the observation on 12 volt Lead-Acid battery (MF40Sv) used in cars. Various batches of data are applied as input parameters to the ANN, viz : Internal Resistance (1), Terminal Voltage (2), Extracted Battery Temperature (3), Current Drawn from battery (4), and Current Consumption Time (5). These five preferred parameters of data when applied as input parameters to the trained ANN, provide indirect indication of Specific Gravity of the battery. Measurement of the various parameters of the battery is made in ambient temperature and operating conditions under which the battery is going to be used, that is, under onload condition of the battery. The temperature-IR characteristics of the battery is drawn by measuring battery Temperature and the Internal Resistance at three to four operating temperatures with different ages of battery at various levels of discharge conditions, where the operating temperature is the environment temperature, the age of the battery refers to the age from the date of manufacturing of the battery, the levels of discharge condition refers to different charged states of the battery under test. Voltage across terminals of battery and Current drawn from battery are measured through voltage and current measuring devices. Impedance measurement is done in the frequency range lkHz-50kHz with high frequency analyzer impedance measuring device, while Current Consumption time of Battery is monitored through a counter. The battery Temperature input to ANN is extracted by least square fit of Temperature (temp)-Internal Resistance (IR) characteristics. ANN is subjected to a learning procedure with aforesaid observed inputs and specific gravity output. The ANN is implemented and trained using ANN toolbox of MATLAB, which is a very standard scientific and mathematical tool given by Mathworks Inc. USA. Neural Network toolbox is one among the other application specific solutions provided by MATLAB. ANN models the nonlinear relationship of the input parameters with Specific Gravity. The ANN structure is made of multilayer perceptron architecture. In a preferred embodiment, as illustrated in Fig. 2, ANN comprises an input layer formed of five sensory neuron cells namely Internal Resistance, Terminal Voltage, Extracted Temperature, Current Drawn, and Consumption Time, two hidden layers (first and second layers) of 11 neurons in each, and one output neuron for Specific Gravity as output. ANN architecture uses back propagation learning algorithm for batch training using a MATLAB batch training function traingdm (Batch Gradient descent with momentum) with learning rate of value 0.05 and transfer function as Log Sigmoid (logsig) and linear (purelin). Back propagation algorithm is based on error correction learning rule in a supervised manner, The training function 'traingdm' provides faster convergence with momentum. Acting like a low pass filter, momentum allows the network to ignore small features in error surface. Learning rate which is the rate of size of weights adjustments made in each iteration of learning of ANN, influences the rate of convergence of the learning algorithm. Transfer functions of ANN architecture are the activation functions of neuron of hidden and output layers each, i.e. sigmoid for hidden layer and linear for output layer. where Wj is the synaptic coefficients or weights of connection between jth and ith neuron, Xj is the input excitation to ith neuron and Bj is bias at jth neuron. The neuron produces the weighted sum denoted T and gives an intermediate output signal. Said intermediate output signal is combined with the non-linear activation functions at the hidden layers. Standard transfer function as Log Sigmoid, depicted in Fig. 3, has been employed at both hidden layers: Such functions are preferred for realizing the activating functions in the neurdns of the hidden layer. The activation function for the output layer is selected to be linear function, and the equation formed at the output layer is : where v is the output of single neuron and a is constant. The training of ANN is done with data observed at various external parameter values such as operation temperature 0, 27, and 40 degree Celsius; Real cranking and Slow discharge of car operation; New and aged - 1 year old and 2 years old battery. Specific gravity with applied temperature correction indicates SoC of the battery. A 2-input and 1-output fuzzy logic module is used to determine the SoC of the battery. The SoCof the battery, under test, is determined by using a fuzzy model (fuzzyfier) taking the input Specific Gravity as obtained from ANN and extracted Temperature of the battery. The SoC of the battery is represented in terms of five linguistic variables using fuzzy rule base. MATLAB toolbox 'Fuzzy Logic' is employed for implementing the fuzzy rulebase. The relationship between the mapped Specific Gravity through ANN and battery charged status (SoC) is modeled by said fuzzyfier at corresponding extracted Temperature levels, using the fuzzy rule base, whereby the charged state of the battery is represented linguistically in terms of fuzzy set definition. Fuzzy set definition defines the membership functions of the terms very very high (WH), very high (VH), high (H), medium (M), low (L), very low (VL), very very low (WL) in respect of specific gravity (Fig, 4B); very high (VH), high (H), medium (M), low (L), very low (VL) in respect of extracted Temperature (Fig. 4A); and for linguistic variables of Charged State of battery (Fig. 4C). Membership function defines the closeness of degree by which a member belongs to a fuzzy set, where a fuzzy set is a set which contains elements that have varying degrees of membership in the set. Table 1 shows the Rule Specification of Fuzzy indicators i.e. the fuzzy rulebase employed by the Fuzzy model for the determination of the State of Charge of a battery. The State of Charge of the battery is represented linguistically as 'fully charged', 'more than half charged', 'half charged', 'less than half charged', and 'flat charged'. The built-in membership function types from the Fuzzy Logic Toolbox of MATLAB, used in an illustrative embodiment of the invention, are Gaussian Distribution Curve Based Membership Functions (Gaussmf, gbellmf) and Polynomial Curve Based Membership Functions (zmf, smf). The fuzzyfier employs fuzzy rule base as shown in Table 1, with the membership function as given below. The membership function chosen to represent the extracted temperature and Specific Gravity boundaries, is Gaussian function with property where the parameter c is the center of gravity of Gaussian function and is the distance of the point from c where degree of membership function is half. The left open membership function for the Temperature and Specific Gravity is defined as : where the parameters a and b are extremes of the function. The membership functions for the state of charge are defined as follows : The parameters used for deploying various fuzzy characteristics of each of the aforesaid fuzzy variables Temperature, Specific Gravity and State of Charge are given below: INPUT 1: TEMPERATURE (referring to Figure 4A) The State of Health (SoH) of the battery is determined using a mathematical model. To obtain the online SoH of battery in terms of percentage remaining life of the battery, a mathematical relationship is drawn between the parameters Current Drawn (4), Current Consumption Time (5), Total Consumption Time (6) and the slope of Specific Gravity to Total Consumption Time, where the Total Consumption Time is the time of total battery consumption, that is the time corresponding to different levels of discharge conditions of the battery. A function for percentage of life remaining is drawn using slope of Specific Gravity to the Total Consumption Time, Current Drawn from battery, and the Consumption Time. The mathematical relation employed to estimate the SoH of battery is as follows: where TT = Total Consumption Time, C = Current Drawn, CT = Current Consumption Time, m = slope of Specific Gravity to Total Consumption Time, and a, b are coefficients. The measuring devices employed for online input to ANN are : Temperature : This is extracted from the IR-temperature characteristic of the battery in question. Current : This is measured with the basic current probe. It includes switching implemented to work in two ranges, viz., (i) 10-20 amperes in case of slow discharge, and (ii) 100-200 amperes in case of real cranking. Terminal voltage : This is measured using a simple voltmeter. Internal resistance : This is measured with an oscillator in the frequency range lkHz-to 50 kHz with Frequency Analyzer Circuit. Consumption time : This is counted using a counter which resets itself whenever the discharging ends. This is implemented through the Microcontroller. In addition, a power back-up is used for the microcontroller and associated circuits. The membership function must vary between 0 and 1. The function itself can be an arbitrary curve whose shape one can define as a function that suits the application from the point of view of simplicity, convenience, speed and efficiency. The Fuzzy Logic Toolbox includes 11 built-in membership function types : trimf, trapmf, gaussmf, gauss2mf, gbellmf, sigmf, dsigmf, psigmf, zmf, pimf, smf. The present invention uses • Gaussian distribution curve based membership functions Gaussmf : gaussian curve based gbellmf: Bell shaped membership function • Polynomial curves based membership functions Zmf, Smf : Z, S curves are all so named because of their shaDes. The function zmf is the asymmetrical polynomial curve open to the left and smf is the mirror-image function that opens to the right. The Fuzzy Logic Toolbox also allows the user to create one's own membership functions. The study of training functions in Artificial Neural Network has been done using two training functions from MATLAB: > trainbr > traingdm Trainbr is a batch training function wherein the weight and biases of the network are updated only after entire training set has been applied to the network. Traingdm is also a batch training function but with steepest descent with momentum. It provides faster convergence. Momentum allows a network to respond not only to the local gradient, but also to recent trends in error surface. Acting like a low pass filter, momentum allows the network to ignore small features in error surface. Without • momentum, a network may get stuck in shallow local minimum. The process according to the present invention can be implemented as a product Dy a number of various instrument architectures varying in cost and complexity. In one such low cost implementation schematics, a Micro-controller having NVRAM stores the ANN architecture, weights and the fuzzy rule base and flash RAM for temporary usage. The implementation is general. The selection of trie micro-controller for implementing the ANN and the fuzzifier, having NVRAM and flash memory to store ANN weights and fuzzy rule-base, is such that once implemented, the design becomes battery specific. The analysis of intelligent battery monitoring process uses five preferred parameters for ANN. In the process according to the present invention, complex and non-linear behavior of battery has been mapped through training of ANN due to its in-bult capacity to handle the non-linearity efficiently. The indication of SoC and SoH of battery depends on large number of parameters, most of which are impossible to detect under onlqaa condition or without support of sufficient laboratory instruments. Specific gravity of battery is a parameter which can be monitored through a chemical process and it is difficult to get online. Hence, five preferred parameters have been used which can provide the Indirect indication of specific gravity, corrosion, etc., and in turn, SoC and SoH. The preferred five parameters are terminal voltage, current consumption, incernat resistance of battery, temperature and consumption time. The preferred parameters provide indirect indication of Specific Gravity when given to trained ANN as input. The Specific Gravity with applied temperature correction indicates SoC. While the invention has been described using five preferred parameters, those skilled in the art will appreciate that the experiment can be conducted with three or four essential parameters, even though a less accurate model may be obtained. This can be better understood from Table 3 which shows the comparative analysis with three or four essential input parameters and five preferred input parameters applied to ANN and the corresponding experimental results obtained thereby. Reference numerals 1, 2, 3, 4 and 5 indicating the various inputs to ANN, in Table 3, refer to the following parameters : 1- Terminal voltage 2- Consumption current 3- Internal resistance 4- Temperature 5- Consumption time Epochs are the cycles required to train the ANN for given target, and surface error is the root mean square value of error which indicates the closeness of training of ANN model. The experimental results in the table show that ANN can be trained to generate the result, using three or four essential input parameters. However, the most accurate ANN training can be achieved with the help of atleast five preferred input parameters, in which case ANN succeeds in giving an output which is more close to the real value. The model works successfully at environment temperatures ranging from 0-50 degree Celsius in case of Lead Acid batteries used in cars. However, the model, as described hereinbefore, is general and hence, can be used in varying operating conditions and for different types of batteries. The foregoing is to be considered as illustrative only of the principles of the present invention. Depending on the battery under test, the measured input-output values to train ANN and the rule base and values used in membership function for rule base application will vary and can be experimented with. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the invention to the exact construction and applications shown and described, and accordingly, ail suitable modifications and equivalents may be resorted to, falling within, the scope of the invention in the appended claims and their equivalents. WE CLAIM: 1. A process of determining State of Charge and State of Health of a battery under on- load conditions, said process comprising the steps of: drawing the temperature-internal resistance characteristics of a battery by measuring the battery temperature and the internal resistance at three to four operating temperatures with different ages of battery at various levels of discharge conditions; training an artificial neural network (ANN) with measured inputs, namely, terminal voltage, current drawn from the battery and internal resistance, and, optionally, with other inputs, duly measured, and selected from one or more of battery temperature and consumption time, for the purpose of obtaining the output Specific Gravity of the battery, said battery temperature input to the ANN being extracted by least square fit of said (temp-IR) characteristics; determining the State of Charge of the battery by deploying fuzzy rule base using the battery temperature, as extracted, and the Specific Gravity, as found in the preceding step; and determining the State of Health of the battery in terms of percentage of remaining life of the battery, based on the measured inputs such as, current drawn from the battery, consumption time, total consumption time and slope of said Specific Gravity to total consumption time, and using a mathematical model, such as herein described. 2. A process as claimed in claim 1, wherein the operating temperature of the battery is the ambient temperature. 3. A process as claimed in any of claims 1 and 2, wherein the age of the battery is the age from the date of its manufacture. 4. A process as claimed in any of the preceding claims, wherein said battery is a 12v Lead-Acid battery of the type used in cars, and said State of Charge of the battery is determined in terms of five linguistic variables, namely, fully charged, more than half charged, half charged, less than half charged, flat charged. 5. A process as claimed in any preceding claim, wherein the consumption time is the current consumption time of battery and total consumption time is the time of total battery consumption time corresponding to different levels of discharge conditions which relate to the age of battery. 6. A process as claimed in any preceding claim, wherein the mathematical model used for determining State of Health of a battery under on-load condition is based on the following mathematical relation: SoH = [{TT*C - (a*C*CT + b*m)}/TT]*100 where TT = Total Consumption Time, C = Current Drawn, CT = Current Consumption Time, m = slope of Specific Gravity to Total Consumption Time, and a, b are coefficients. 7. A process as claimed in any preceding claim, wherein the ANN architecture is a multilayer perceptron architecture having two hidden layers of eleven neurons in each, with five input sensory neurons, namely, internal resistance, terminal voltage, current drawn from the battery, consumption time and extracted temperature, and one output neuron for specific gravity as output. 8. A process as claimed in claim 7, wherein the ANN architecture uses back propagation learning algorithm for batch training using MATLAB function traingdm with learring rate of value 0.05 and standard transfer functions as Log Sigmoid employed at both hidden layers and linear employed at the output layer. 9. A process as claimed in claim 8 wherein the back propagation algorithm is based on error correction learning rule in a supervised manner. 10. A process as claimed in any preceding claim, wherein said fuzzy rule base has relationship of said specific gravity and corresponding extracted temperature with the linguistic variable representing charged state of the battery in terms of fuzzy set definition. 11. A process as claimed in any of the preceding claims, wherein the output of said fuzzy rule base, deployed for determining the state of charge of the battery, is in terms of fuzzy set definition which defines the membership functions of the terms very very high (VVH), very high (VH), high (H), medium (M), low (L), very low (VL), very very low (VVL) for specific gravity, very high (VH), high (H), medium (M), low (L), very low (VL) for extracted temperature, and for linguistic variables of charged state of the battery. |
---|
00813-kol-2005-description complete.pdf
813-KOL-2005-AMENDED CLAIMS.pdf
813-KOL-2005-AMENDED PAGES OF SPECIFICATION.pdf
813-kol-2005-correspondence.pdf
813-kol-2005-description (complete).1.1.pdf
813-kol-2005-examination report reply recieved.1.2.pdf
813-kol-2005-examination report.pdf
813-kol-2005-granted-abstract.pdf
813-kol-2005-granted-claims.pdf
813-kol-2005-granted-description (complete).pdf
813-kol-2005-granted-drawings.pdf
813-kol-2005-granted-form 1.pdf
813-kol-2005-granted-form 2.pdf
813-kol-2005-granted-specification.pdf
813-KOL-2005-REPLY TO EXAMINATION REPORT.pdf
813-kol-2005-reply to examination report1.1.pdf
Patent Number | 246616 | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Indian Patent Application Number | 813/KOL/2005 | ||||||||||||
PG Journal Number | 10/2011 | ||||||||||||
Publication Date | 11-Mar-2011 | ||||||||||||
Grant Date | 07-Mar-2011 | ||||||||||||
Date of Filing | 05-Sep-2005 | ||||||||||||
Name of Patentee | EXIDE INDUSTRIES LIMITED | ||||||||||||
Applicant Address | 59E, CHOWRINGEE ROAD, CALCUTTA 700020, WEST BENGAL, INDIA | ||||||||||||
Inventors:
|
|||||||||||||
PCT International Classification Number | H01M 10/46 | ||||||||||||
PCT International Application Number | N/A | ||||||||||||
PCT International Filing date | |||||||||||||
PCT Conventions:
|