DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
The Amendment filed 3/11/2026 has been entered. Claims 1-4, 6-12, and 14-15 remain pending in the application with added new claim 15, and claims 5 and 13 have been canceled. Applicant’s amendments to the Claims have overcome every 103 rejection previously set forth in the Non-Final Office Action mailed 12/18/2025. The new grounds of rejection presented below are necessitated by the amendments. Accordingly, this Office Action is made Final.
Response to Arguments
Applicant's arguments filed 3/11/2026 have been fully considered but they are not persuasive.
Applicant submits on pages 7-8 of Remarks that Mohajer and Juang do not teach “wherein using the deterministic gradient method includes determining a gradient of the loss function at maximum and minimum outer limit values of the loss function f(d), and at a mean value of the loss function f(d),” as recited in claim 1.
The applicant points to paragraph [32] of Juang which recites
"The RNN is a type of multi-layered neural network that retains memory from the previous iteration. Such internal memory helps process arbitrary sequences of inputs (e.g., current, temperature, amp-hour), to output dynamic temporal behavior of the battery (e.g., voltage). The RNN integrates coherently with the backpropagation training method and the stochastic gradient descent optimization method, to learn the appropriate internal representations of a mapping of input to output. Compared to the ECM and the physics based models, a RNN captures the battery nonlinearity, takes the temperature dependency into account, and does not require foreknowledge of battery physical parameters."
highlighting that Juang uses a training procedure relying on stochastic gradient descent, which performs parameter updates using randomly sampled subsets of data, producing inherently probabilistic, non-repeatable optimization trajectories tied to the statistical properties of the training dataset. In other words, Juang's random minibatch selection means its gradient estimates vary from iteration to iteration, causing the optimization path and often even the final parameter values-to differ across training runs. The cited portions of Juang fail to disclose deterministic gradient method. In contrast, claim 1 generally recites a deterministic gradient method applied to a single-parameter, closed-form loss function f(d) that is explicitly defined, parabolic, and evaluated at predetermined parameter values (e.g., maximum, minimum outer limit values and a mean value of a loss function) to locate a unique global minimum with no stochasticity, random sampling, or data-dependent variability. Accordingly, Juang does not teach, disclose, or suggest at least the following subject matter: "wherein using the deterministic gradient method includes determining a gradient of the loss function at maximum and minimum outer limit values of the loss function f(d), and at a mean value of the loss function f(d)" as recited in claim 1 as amended.
The examiner submits that under the broadest reasonable interpretation, Mohajer, in view of Juang and as evidence by Brownlee, discloses the limitation, the use of a deterministic gradient method to determine the minimum of a loss function of a parameter (Juang - ¶’s [31-32]: Recurrent Neural Networks (RNN) for battery modeling uses stochastic gradient descent method for appropriate mapping of input to output based on temperature dependency and does not require foreknowledge of battery physical parameters. The examiner interprets the loss function as a loss of optimization function where the parameters are adjusted to minimize optimization loss. The examiner interprets term “deterministic” applied to the gradient method as the process of determining the minimum of optimization loss),
wherein using the deterministic gradient method includes determining the gradient of the loss function at maximum and minimum outer limit values of the loss function f(d), and at a mean value of the loss function f(d) (As evidenced by Brownlee page 5/18, the stochastic gradient descent method determines coefficients for the optimization function to make a best fit for a trendline against data. Through the process, maximum and minimum out limits of the errors are determined and errors are reduced to be closer to zero over more iterations).
The explanation of “deterministic gradient method,” being allegedly is distinct from the “stochastic gradient descent” of Juang, given by the applicant does not appear to be supported by claim language or the Specification. The closest concept the examiner can find are in paragraphs [14, 25, 52, 65-66] with phrases such as “a minimum of a loss function f(d) of a parameter d of a polynomial charging profile Ip(t) is numerically determined”] and “the gradient of the loss function f(d) at the outer limit values dmin and dmax of the loss function f(d) and at a mean value dm of the loss function f(d) halfway between the outer limit values dmin and dmax is first of all determined.”
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-4, 6-12, and 14-15 are rejected under 35 U.S.C. 103 as being unpatentable over the ScienceDirect Publication of Mohajer et al. (“Design of a Model-based Fractional-Order Controller for Optimal Charging of Batteries,” IFAC Conference PapersOnLine, ScienceDirect, presented Sep. 4-6, 2018) in view of Juang et al. (US 20180086222 A1) as evidenced by Brownlee (“Linear Regression Tutorial Using Gradient Descent for Machine Learning” Machine Learning Mastery, < https://machinelearningmastery.com/linear-regression-tutorial-using-gradient-descent-for-machine-learning/ > Published online 3/29/2016), hereinafter respectively referred to as Mohajer, Juang, and Brownlee.
Regarding independent claim 1, Mohajer teaches a charger (Fig. 6 and Section 4: robust controller) for an electrical energy store (Fig. 6: battery),
wherein the charger has an open-loop control unit (plant model current trajectories, Ich trajectories and Jsr trajectories) and a closed-loop control unit (controller C(s)),
wherein the charger is configured to charge the electrical energy store to a defined state of charge within a preset charging time (Fig. 2: fixed charging time of 20 minutes) and to set a charging current (Fig. 6: ICh,traj) and a side reaction current (Jsr,traj or Jsr,obs) of the electrical energy store (p. 98, col. 1, par. 1-2, Sec. 2; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature)
wherein an affine or polynomial charging profile is optimized by numerically determining a minimum of a loss function of a parameter of the charging profile, wherein a first charging current and a first side reaction current are subjected to open-loop control according to an optimized charging profile (p. 98, col. 1, par. 1-2, Sec. 2; p. 100, col. 2, par. 1; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature. The examiner interprets the loss function as a loss of optimization function where the parameters of SOC, aging, and temperature are adjusted to minimize optimization loss ),
wherein the charging current is one of a first charging current (ICh,traj), a second charging current (ICh,FF), a third charging current (ICh,FB), or a fourth charging current (Icharge),
wherein the side reaction current is one of a first side reaction current (Jsr,traj) or a second side reaction current (Jsr,obs),
Mohajer does not explicitly teach the use of a deterministic gradient method to determine the minimum of a loss function of a parameter of the charging profile,
wherein using the gradient method includes determining the deterministic gradient of the loss function at maximum and minimum outer limit values of the loss function f(d), and at a mean value of the loss function f(d).
Juang teaches the use of a deterministic gradient method to determine the minimum of a loss function of a parameter (¶’s [31-32]: Recurrent Neural Networks (RNN) for battery modeling uses stochastic gradient descent method for appropriate mapping of input to output based on temperature dependency and does not require foreknowledge of battery physical parameters. The examiner interprets the loss function as a loss of optimization function where the parameters are adjusted to minimize optimization loss. The examiner interprets term “deterministic” applied to the gradient method as the process of determining the minimum of optimization loss),
wherein using the deterministic gradient method includes determining the gradient of the loss function at maximum and minimum outer limit values of the loss function f(d), and at a mean value of the loss function f(d) (As evidenced by Brownlee page 5/18, see plot of error versus iteration below, the stochastic gradient descent method determines coefficients for the optimization function to make a best fit for a trendline against data. Through the process, maximum and minimum out limits of the errors are determined and errors are reduced to be closer to zero over more iterations).
PNG
media_image1.png
436
635
media_image1.png
Greyscale
Both Mohajer and Juang teach systems for optimization. Mohajer optimizes charging profiles based on battery SOC, aging, and temperature. It would have been obvious to a person having ordinary skill in the art before the effective filing date of the instant application to incorporate the stochastic gradient descent optimization method in the system of Juang into the system of Mohajer. Doing so would aid Mohajer in the minimization of ageing with optimization of charging based on a given data points in a successive iteration of instances, a simpler way for optimization as compared to depending on a long sequence of data (Juang - ¶0031: long sequence of output currents).
Regarding claim 2, Mohajer teaches the charger as claimed in claim 1, wherein the charger has an evaluation unit (Fig. 6: cell observer), which has at least one terminal for a sensor (sensors that produce temperature signal T and cell voltage signal Ucell) of the electrical energy store (battery), wherein the evaluation unit is configured to determine at least aging (aging observer) of the electrical energy store by means of a simplified linear electrothermal aging model of the electrical energy store (p. 98, col. 1, par. 1-2, Sec. 2; p. 100, col. 2, par. 1; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature. Cell observer measures the cell voltage and temperature which is compared with the measured trajectories of side reaction current).
Regarding claim 3, Mohajer teaches the charger as claimed in claim 2,
wherein the evaluation unit is connected in signal-conducting fashion to the open-loop control unit and/or to the closed-loop control unit (Fig. 6: currents, Jsr,traj of the cell observer and Jsr,obs of Jsr trajectories, are compared and sent to controller C(s)).
Regarding claim 4, Mohajer teaches the charger as claimed in claim 1,
wherein the open-loop control unit (Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories) is configured to subject the first charging current (ICh,traj) and the first side reaction current (Jsr,traj) to open-loop control so that the electrical energy store (battery) is charged to the defined state of charge within the preset charging time (p. 98, col. 1, par. 1-2, Sec. 2; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are based on optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature).
Regarding claim 6, Mohajer teaches the charger as claimed in claim 1, wherein the open-loop control unit has a charge open-loop control means configured to subject the first charging current to open-loop control according to the optimized charging profile (p. 98, col. 1, par. 1-2, Sec. 2; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are based on optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature).
Regarding claim 7, Mohajer teaches the charger as claimed in claim 1, wherein the closed-loop control unit (Fig. 6: controller C(s)) is configured to subject a third charging current (ICh,FB) to closed-loop control so that the second side reaction current of the electrical energy store is minimized (p. 98, col. 2, par. 2, trajectories set to minimize capacity loss).
Regarding claim 8, Mohajer teaches the charger as claimed in claim 1, wherein the charger has a summation means (Fig. 6: the circle by two plus signs and receiving charging currents ICh,FF and IChFB), which is arranged between the open-loop control unit (Ich trajectories and low pass filter LPF) and the closed-loop control unit (controller C(s)), on one side, and an output terminal of the charger (the output line where charge current Icharge meets the battery), on another side, wherein the summation means is configured to add the first charging current or the second charging current (ICh,FF) from the open-loop control unit (Ich trajectories and low pass filter LPF) and the third charging current (ICh,FB) from the closed-loop control unit (controller C(s)) and to generate the fourth charging current (Icharge).
Regarding claim 9, Mohajer teaches the charger as claimed in claim 8, wherein the charger has a low-pass filter (Fig. 6: LPF), which is arranged between the open-loop control unit and the summation means.
Regarding claim 10, Mohajer teaches the charger as claimed in claim 8, wherein the charger has a comparison means (Fig. 6: the circle by a plus sign and minus sign and receiving side reaction currents Jsr,traj and Jsr,obs), which is arranged between the open-loop control unit (JST trajectories) and an ageing evaluation means (aging observer in the cell observer), on one side, and the summation means (the circle by two plus signs and receiving ICh,FB and ICh,FF and outputting ICharge), on another side, wherein the comparison means is configured to compare the first side reaction current and the second side reaction current (Fig. 6: the circle by a plus sign and minus sign and receiving side reaction currents Jsr,traj and Jsr,obs).
Regarding independent claim 11, Mohajer teaches a method for charging an electrical energy store (Fig. 6: battery) by means of a charger having an open-loop control unit (Ich trajectories and Jsr trajectories) and a closed-loop control unit (controller C(s)),
wherein the charger is configured to charge the electrical energy store to a defined state of charge within a preset charging time and to set a charging current (ICh,traj) and a side reaction current (Jsr,traj) of the electrical energy store (p. 98, col. 1, par. 1-2, Sec. 2; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are based on optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature),
wherein the method comprises an open-loop control step and a closed-loop control step, which run simultaneously (Fig. 6: charging current Ich,FF from Ich trajectories and low pass filter LPF is produced independently from charging current Ich,FB from controller (C(s)),
wherein the electrical energy store is charged to a defined state of charge within a preset charging time and a charging current (ICh,traj) and a side reaction current (Jsr,traj) of the electrical energy store are set (p. 98, col. 1, par. 1-2, Sec. 2; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are based on optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature)
wherein an affine or polynomial charging profile is optimized by numerically determining a minimum of a loss function of a parameter of the charging profile, wherein a first charging current and a first side reaction current are subjected to open-loop control according to an optimized charging profile (p. 98, col. 1, par. 1-2, Sec. 2; p. 100, col. 2, par. 1; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature. The examiner interprets the loss function as a loss of optimization function where the parameters of SOC, aging, and temperature are adjusted to minimize optimization loss),
wherein the charging current is one of a first charging current (ICh,traj), a second charging current (ICh,FF), a third charging current (ICh,FB), or a fourth charging current (Icharge),
wherein the side reaction current is one of a first side reaction current (Jsr,traj) or a second side reaction current (Jsr,obs),
Mohajer does not explicitly teach the use of a deterministic gradient method to determine the minimum of a loss function of a parameter of the charging profile,
wherein using the deterministic gradient method includes determining the gradient of the loss function at maximum and minimum outer limit values of the loss function f(d), and at a mean value of the loss function f(d).
Juang teaches the use of a deterministic gradient method to determine the minimum of a loss function of a parameter (¶’s [31-32]: Recurrent Neural Networks (RNN) for battery modeling uses stochastic gradient descent method for appropriate mapping of input to output based on temperature dependency and does not require foreknowledge of battery physical parameters. The examiner interprets the loss function as a loss of optimization function where the parameters are adjusted to minimize optimization loss),
wherein using the deterministic gradient method includes determining the gradient of the loss function at maximum and minimum outer limit values of the loss function f(d), and at a mean value of the loss function f(d) (As evidenced by Brownlee page 5/18, see plot of error versus iteration below, the stochastic gradient descent method determines coefficients for the optimization function to make a best fit for a trendline against data. Through the process, maximum and minimum out limits of the errors are determined and errors are reduced to be closer to zero over more iterations).
PNG
media_image1.png
436
635
media_image1.png
Greyscale
Both Mohajer and Juang teach systems for optimization. Mohajer optimizes charging profiles based on battery SOC, aging, and temperature. It would have been obvious to a person having ordinary skill in the art before the effective filing date of the instant application to incorporate the stochastic gradient descent optimization method in the system of Juang into the system of Mohajer. Doing so would aid Mohajer in the minimization of ageing with optimization of charging based on a given data points in a successive iteration of instances, a simpler way for optimization as compared to depending on a long sequence of data (Juang - ¶0031: long sequence of output currents).
Regarding claim 12, Mohajer teaches the method as claimed in claim 11,
wherein a present state of charge and/or a present state of health and/or the second side reaction current are determined from sensor data of the electrical energy store by means of a simplified linear electrothermal aging model of the electrical energy store (p. 98, col. 1, par. 1-2, Sec. 2; p. 100, col. 2, par. 1; Fig. 2; and Fig. 6: plant model current trajectories, Ich trajectories and Jsr trajectories, are optimized charging profiles over a set period of time that include side reaction current data based on battery SOC, aging, and temperature. Cell observer measures the cell voltage and temperature which is compared with the measured trajectories of side reaction current).
Regarding claim 14, Mohajer teaches the method as claimed in claim 13,
wherein the first side reaction current is compared with the second side reaction current (Fig. 6: the circle by a plus sign and minus sign and receiving side reaction currents Jsr,traj and Jsr,obs), and a third charging current is generated (Ich,FB),
wherein the third charging current is equal to zero when the first side reaction current has the same value as the second side reaction current and/or wherein, when the first side reaction current and the second side reaction current have different values (it is predictable to a person of ordinary skill in the art that the difference between two same currents would yield zero current and the difference between two difference currents would yield a nonzero current value), the third charging current (Ich,FB) is determined so an ageing of the electrical energy store is minimized (p. 98, col. 2, par. 2, trajectories set to minimize capacity loss),
wherein the third charging current (Ich,FB) and the second charging current are added, and a fourth charging current is generated (the circle by two plus signs and receiving charging currents, ICh,FB and ICh,FF , and outputting ICharge), and
wherein the electrical energy store is charged with the fourth charging current (ICharge is sent to the battery).
Regarding claim 15, Mohajer teaches the charger as claimed in claim 1, wherein using the deterministic gradient method further includes:
determining a range, within the maximum and minimum outer limit values, for the parameter (d) based on the mean value of the loss function f(d) and a sign change of the gradient of the loss function (As evidence by Brownlee, pages 3/18 to 4/18, a range is constructed in a gradient method by starting with plugging in a first try value to get the first error value and then proceeding to plug in more values to get a sufficient set of errors to identify the loss function behavior. Lowest and highest input values would correspond to a minimum and maximum of the range. If the try values result in the crossing of the minimum of the loss function (minimum of resulting errors), a sign change for the gradient will occur because the function is changing between a decreasing slope and an increasing slope).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
GeeksforGeeks (“ML – Stochastic Gradient Descent (SGD)” < https://www.geeksforgeeks.org/machine-learning/ml-stochastic-gradient-descent-sgd/ > Published online 9/30/2025)
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Ryu-Sung P. Weinmann whose telephone number is (703)756-5964. The examiner can normally be reached Monday-Friday 9am-5pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Julian Huffman, can be reached at (571) 272-2147. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Ryu-Sung P. Weinmann/Examiner, Art Unit 2859 March 26, 2026
/JULIAN D HUFFMAN/Supervisory Patent Examiner, Art Unit 2859