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 .
Priority
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been filed in parent Application No. CN2023101930952, filed on 02/28/2023.
Status of Claims
Claims 1-3, 5, 9, 10, 12 are amended. Claims 4, 6, 11 are cancelled. Claims 1 – 3, 5, 7, 9, 10, 12 are pending and examined herein.
Claims 1 – 3, 5, 7, 9, 10, 12 are rejected under 35 U.S.C. 103.
Response to Amendment
The amendment filed June 18th, 2026 has been entered. Claims 1-3, 5, 9, 10, 12 are amended. Claims 4, 6, 11 are cancelled. Claims 1 – 3, 5, 7, 9, 10, 12 are pending and examined herein. Applicant’s amendments to the claims have overcome claim objections and 112(b) rejection previously set forth in the Non-Final Rejection Office Action mailed March 19th, 2026. Although applicant amended the abstract of the disclosure, the objection is maintained.
Response to Arguments
Applicant's arguments filed June 18th, 2026 regarding the 35 U.S.C. 101 rejection of claims 1-7,9-12 have been considered and are persuasive. The 35 U.S.C. 101 rejection of claims 1-7,9-12 has been withdrawn.
Applicant's arguments filed June 18th, 2026 regarding the 35 U.S.C. 103 rejection of claims 1-7,9-12 have been considered but are not persuasive.
Applicant argues that Singleton uses the extended Kalman filter (EKF) only to extrapolate remaining useful life and does not use the EKF as recited in claim 1. However, Singleton defines an EKF state vector containing the exponential degradation function and its unknown parameters a_k and b_k, updates those parameters when new measurements are received, and calculates an updated state estimate. Singleton therefore teaches using the EKF to estimate the exponential degradation model parameters and obtain a state estimate. Singleton also teaches obtaining time dependent degradation models through curve fitting and identifies an exponential function as the most suitable model for the variance feature. The rejection does not rely on Singleton alone for inputting that state estimate into a neural network.
Applicant further argues that Ma and Wang are data driven approaches, whereas Singleton and Gang Wang are model based approaches. The rejection is based on their combined teachings. Singleton supplies the EKF based state estimate while Wang supplies FFT derived frequency domain vibration data used as neural network training input. It would have been obvious to use both inputs because the state estimate represents the estimated degradation condition of the component and the frequency domain data represents measured vibration characteristics associated with that condition. Using both forms of information would predictably provide additional degradation information for training the remaining useful life (RUL) prediction model.
Applicant also argues that Gang Wang uses FPT to initiate RUL prediction rather than to determine training data. Gang Wang teaches that the data after FPT correspond to the fault or degradation stage and that RUL prediction is performed during that stage. In view of Wang’s teaching of applying FFT to bearing vibration data to generate neural network inputs, it would have been obvious to apply Wang’s FFT processing to the post FPT degradation stage data identified by Gang Wang
The newly added limitations do not overcome the rejection. Wang teaches acquiring bearing vibration signals using horizontal and vertical acceleration sensors. Singleton teaches updating its degradation model parameters and state estimate when new measurements are received. Wang teaches applying FFT to test set vibration signals and Ma teaches preprocessing real time sensor data and inputting the processed data into a trained RUL model during online testing. It would have been obvious to apply the EKF and FFT processing to newly acquired monitoring data and input the resulting data into the trained model so that the model receives the same types of degradation information during prediction that were used during training.
Applicant has not presented a separate argument for dependent claims apart from the limitations incorporated from claim 1. Accordingly, the rejection of claims 1 – 3, 5, 7, 9, 10, 12 are maintained under 35 U.S.C. 103.
Specification
Applicant is reminded of the proper content of an abstract of the disclosure.
A patent abstract is a concise statement of the technical disclosure of the patent and should include that which is new in the art to which the invention pertains. The abstract should not refer to purported merits or speculative applications of the invention and should not compare the invention with the prior art.
If the patent is of a basic nature, the entire technical disclosure may be new in the art, and the abstract should be directed to the entire disclosure. If the patent is in the nature of an improvement in an old apparatus, process, product, or composition, the abstract should include the technical disclosure of the improvement. The abstract should also mention by way of example any preferred modifications or alternatives.
Where applicable, the abstract should include the following: (1) if a machine or apparatus, its organization and operation; (2) if an article, its method of making; (3) if a chemical compound, its identity and use; (4) if a mixture, its ingredients; (5) if a process, the steps.
Extensive mechanical and design details of an apparatus should not be included in the abstract. The abstract should be in narrative form and generally limited to a single paragraph within the range of 50 to 150 words in length.
See MPEP § 608.01(b) for guidelines for the preparation of patent abstracts.
Applicant is reminded of the proper language and format for an abstract of the disclosure.
The abstract should be in narrative form and generally limited to a single paragraph on a separate sheet within the range of 50 to 150 words in length. The abstract should describe the disclosure sufficiently to assist readers in deciding whether there is a need for consulting the full patent text for details.
The language should be clear and concise and should not repeat information given in the title. It should avoid using phrases which can be implied, such as, “The disclosure concerns,” “The disclosure defined by this invention,” “The disclosure describes,” etc. In addition, the form and legal phraseology often used in patent claims, such as “means” and “said,” should be avoided.
The abstract of the disclosure is objected to because the opening language substantially repeats the title of the invention. Also, the phrases “by means of” should be replaced with clearer narrative language, such as “using.” As recited above, the abstract should be clear and concise and should not repeat information given in the title and should avoid using legal phraseology which are often used in claims. A corrected abstract of the disclosure is required and must be presented on a separate sheet, apart from any other text. See MPEP § 608.01(b).
Claim Objections
Claims 1, 2, 9 are objected to because of the following informalities:
Claims 1, 2, 9 recite “based on the the degradation model” which contains the duplicated word “the”.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1 – 3, 5, 7, 9, 10, 12 are rejected under 35 U.S.C. 103 as being unpatentable over Singleton et al. (NPL: “Extended Kalman Filtering for Remaining-Useful-Life Estimation of Bearings”) in view of Ma et al. (U.S. Pub. 2025/0021089 A1), Wang et al. (foreign pub. CN109726524 B), further in view of Gang Wang et al. (NPL: “Remain useful life prediction of rolling bearings based on exponential model optimized by gradient method”).
Regarding Claim 1, Singleton teaches
modeling a degradation process of the mechanical component by fitting a variance feature sequence extracted from the original state monitoring data through an exponential random model , to obtain a degradation model of the mechanical component, and establishing a system state space equation based on the the degradation model of the mechanical component; (Pg. 1785 C.EKF Parameter Learning section of Singleton states “For the two different types of features, i.e., vibration and entropy, time-dependent degradation models are obtained through curve fitting. For the variance feature, an exponential of the form aebt was found to be the most suitable, whereas for the TF entropy feature, a curve in the form of a − be−ct was more suitable” Singleton teaches fitting a sequence of variance features obtained from bearing vibration data with an exponential degradation model. Pg. 1783 D. Extended Kalman Filtering section of Singleton states “The EKF equation in the presence of process noise and measurement noise is xk = f(xk−1, uk−1) +wk−1 (3) where xk is the state being estimated, f is a nonlinear function of states, uk is the input at time sample k, wk is random zero mean noise with covariance matrix Qk. In EKF, the relationship between system states (xk) and measurements (zk) can also be nonlinear, i.e.,zk = h(xk)+vk (4) where zk is the measurement, h is a measurement function that is a nonlinear function of states, and vk is a zero-mean random process” Singleton establishes a state space representation based on the exponential degradation model. The process error and measurement error supply the stochastic or random component of the state space model. )
estimating, based on the system state space equation, parameters of the exponential random model by an extended Kalman filter, to obtain an optimal state estimate; (Pg. 1785 C.EKF Parameter Learning section of Singleton states “The parameters of the degradation function are updated with each new measurement. To accomplish this, a state vector x containing the equation for the curve fit as well as the unknown parameters describing this degradation model at each time point are defined for each feature. For the variance feature, the parameters ak and bk of the exponential curve are used to define the state vector as [43], [53] xk = [akebkk ak bk]T (5)… With each time step, the parameters of the degradation model are updated to form a new model, i.e., fk, and an estimate of the next state, i.e., ˆxk, is calculated…. An overall view of the algorithm and all its steps are given below: 1) Initialize x0 and P0. 2) Predict the next state, ˆxk, and uncertainty matrix, Mk, i.e., ˆxk =f(ˆxk−1) + wk−1 (14) Mk =Fk−1Pk−1Fk−1 + Qk−1. (15) 3) Take in measurement zk. 4) Update the predictions and their uncertainties using the Kalman gain, Kk, ” Since the EKF state vector contains the exponential model value and parameters, Singleton’s updated EKF state estimate corresponds to the claimed optimal state estimate under BRI. Optimal is mapped to the estimate obtained after the EKF prediction and measurement update operations )
according to the optimal state estimate and (Pg. 1785 C.EKF Parameter Learning section of Singleton states “With each time step, the parameters of the degradation model are updated to form a new model, i.e., fk, and an estimate of the next state, i.e., ˆxk, is calculated….”)
However, Singleton does not explicitly teach that
acquiring original state monitoring data of the mechanical component, wherein the original state monitoring data comprises a plurality of vibration signals monitored by sensors in both a horizontal direction and a vertical direction;
determining a first predicting time based on the original state monitoring data of the mechanical component; wherein the first predicting time divides a life cycle of the mechanical component into a health stage and a degradation stage;
extracting, from the original state monitoring data of the mechanical component, state monitoring data in the degradation stage of the mechanical component based on the first predicting time, and extracting, by using fast Fourier transform (FFT), frequency domain data corresponding to the state monitoring data in the degradation stage;
constructing a neural network training data set of all mechanical components according … the frequency domain data;
building a hybrid driven prediction model comprising a fully connected layer, a one-dimensional convolutional long short-term memory network adaptive encoding layer, a multi-head attention mechanism module, a feedforward module and a fully connected regression layer;
training and testing the hybrid driven prediction model by using the neural network training data set, to obtain a trained hybrid driven prediction model;
acquiring new state monitoring data of the mechanical component, processing the new state monitoring data by the extended Kalman filter and the FFT,
and inputting the processed new state monitoring data into the trained hybrid driven prediction model to predict the remaining useful life of the mechanical component
Wang teaches that
acquiring original state monitoring data of the mechanical component, wherein the original state monitoring data comprises a plurality of vibration signals monitored by sensors in both a horizontal direction and a vertical direction; (Pg. 7 of Wang states “the full life data of the rolling bearing acceleration degradation test is from the PRONOSTIA test table[19], respectively collecting data by two acceleration sensors in the horizontal direction and the vertical direction, recording once data every 10 s, the sampling time is 0.1s, the sampling frequency is25.6kHz That is, each sampling vibration data is 2560 points. when the rolling bearing is normally run to the complete failure period, when the acceleration amplitude is continuously more than 20 g, it is considered that the rolling bearing is completely invalid.”)
and extracting, by using fast Fourier transform (FFT), frequency domain data corresponding to the state monitoring data in the degradation stage; (Pg. 4 of Wang states “The method firstly performs FFT conversion to the original vibration signals the rolling bearing, then the frequency domain amplitude signal obtained by the pre-processing is normalized, and it is used as the input of CNN. using CNN with convolution operation, weight sharing and so characteristics, automatically extracting data local abstract information to excavate the deep features, avoiding the problem that the traditional feature extraction method is too dependent on expert experience.” Pg. 7 of Wang states “Generally, the rolling bearing has a performance degradation failure and burst failure mode, the invention combines with the characteristics bearing vibration signals and deep learning and LSTM each advantage, digging deep characteristic and constructing trend quantitative health index, paintable bearing degradation trend and realizing the prediction of the rolling bearing RUL. FIG. 3 is a block diagram of the whole process frame. The specific process steps are as follows: 1) selecting the part data of each working condition of the rolling bearing vibration signals the training set, and performing FFT conversion to the original vibration signals of the training set to obtain the frequency domain amplitude value signal. 5) performing FFT conversion to the non-full-life time domain vibration signals of different working conditions in the test set, obtaining the frequency domain amplitude signal, and performing normalization processing, step 3) digging the deep characteristic, combining the trend quantitative health index model of step 4), obtaining the trend quantitative health index of the test set. using the moving average method (average, MA) to smooth [17], reducing oscillation to improve the remaining life prediction precision. and 6) fitting the performance degradation trend the rolling bearing by the polynomial curve [18] to predict the RUL of the rolling bearing” It would have been obvious to apply Wang’s FFT processing to the post FPT vibration data identified in Gang Wang. It would produce frequency domain data specifically corresponding to the degradation stage.)
constructing a neural network training data set of all mechanical components according … the frequency domain data; (Wang states “1) selecting the part data of each working condition of the rolling bearing vibration signals the training set, and performing FFT conversion to the original vibration signals of the training set to obtain the frequency domain amplitude value signal. 2) the frequency domain amplitude signal after performing normalization processing as characteristic input, life percentage as output training model, formula represents the N-dimensional characteristic at the time t, N=2048, yt belongs to [0, 1] represents the life degradation percentage of the bearing at time t, Dtra represents a certain working condition in the training vibration signals of a bearing of the data, R is frequency domain amplitude characteristic matrix; T is the running time of the bearing full service life. 3) setting number of CNN layers, filter number, convolution size and other parameters, the normalized frequency domain amplitude signal as input of CNN, using the convolution layer in CNN, the formula (1) to (3) of the pool layer traversal the whole input data sequence, extracting the local information of the vibration signals, mining deep features. 4) then inputting the deep characteristic into the LSTM network, using the advantages of the formula (4) to (9) and LSTM memory unit for long term memory to the time sequence data, constructing trend quantitative health index through LSTM network, establishing trend quantitative health index model. 5) performing FFT conversion to the non-full-life time domain vibration signals of different working conditions in the test set, obtaining the frequency domain amplitude signal, and performing normalization processing, step 3) digging the deep characteristic, combining the trend quantitative health index model of step 4), obtaining the trend quantitative health index of the test set. using the moving average method (average, MA) to smooth [17], reducing oscillation to improve the remaining life prediction precision. and 6) fitting the performance degradation trend the rolling bearing by the polynomial curve [18] to predict the RUL of the rolling bearing.” While Singleton teach the model based state estimate degradation parameter input, Wang teaches forming a training set using FFT to obtain frequency domain signal feature for neural network training. It would have been obvious to construct the neural network training data using both Singleton’s EKF state estimate and Wang’s FFT derived features. The state estimate represents the estimated degradation trend and model parameters, while the FFT data represent measured frequency characteristics of the same bearing. Using both time corresponding feature types would predictably provide the neural network with complementary model based and signal based information concerning component degradation. Both are numerical, time indexed features produced from the component’s monitoring measurements so they could have been used together as model inputs. )
a one-dimensional convolutional long short-term memory network adaptive encoding layer, (Wang states “FIG. 1 is a one-dimensional CNN diagram, FIG. 2 is a LSTM memory unit structure diagram.. The invention claims a method for predicting rolling bearing RUL based on convolutional neural network (CNN) and long short-time memory neural network (LSTM) constructing trend quantitative health index, the fast Fourier transform (fast fourier transform, FFT) obtained frequency domain amplitude signal after normalization processing as input of CNN, digging deep features, avoiding the problem that traditional feature extraction method is too dependent on expert experience. then, using LSTM network with good processing time sequence advantage, constructing the trend quantization health index, so as to further predict RUL of rolling bearing, suitable for prediction of performance degradation gradual fault and burst fault two modes.” Wang teaches the convolutional feature extraction and LSTM temporal encoding functions. It would have been obvious to integrate Wang’s one dimensional convolutional processing into the recurrent LSTM transformations so that local feature relationships are retained during each temporal update. This would consolidate the convolutional and temporal processing functions that Wang already applies successively to the same vibration features, with each function continuing to perform its established purpose.)
acquiring new state monitoring data of the mechanical component, processing the new state monitoring data by the extended Kalman filter and the FFT, (Pg. 1785 of Singleton states “For the two different types of features, i.e., vibration and entropy, time-dependent degradation models are obtained through curve fitting. For the variance feature, an exponential of the form aebt was found to be the most suitable, whereas for the TF entropy feature, a curve in the form of a − be−ct was more suitable [see Fig. 7(a) and (b)] The parameters of the degradation function are updated with each new measurement. To accomplish this, a state vector x containing the equation for the curve fit as well as the unknown parameters describing this degradation model at each time point are defined for each feature.” Pg. 3 of Wang states “5) performing FFT conversion to the non-full-life time domain vibration signals of different working conditions in the test set, obtaining the frequency domain amplitude signal, and performing normalization processing, step 3) digging the deep characteristic, combining the trend quantitative health index model of step 4), obtaining the trend quantitative health index of the test set. using the moving average method (average, MA) to smooth [17], reducing oscillation to improve the remaining life prediction precision. and 6) fitting the performance degradation trend the rolling bearing by the polynomial curve [18] to predict the RUL of the rolling bearing.” It would have been obvious to apply Singleton’s EKF processing and Wang’s FFT processing to newly acquired monitoring data because those are the corresponding operations used to generate the model based and signal based features supplied during training.)
Gang Wang teaches that
determining a first predicting time based on the original state monitoring data of the mechanical component; wherein the first predicting time divides a life cycle of the mechanical component into a health stage and a degradation stage; (Pg. 1 Abstract of Gang Wang states “Remaining useful life (RUL) using exponential model (EM) prediction has been a hot research topic in the construction of prognostics health management (PHM) systems. However, in RUL prediction of rolling bearings, the EM 1) depends on the appropriate first prediction time (FPT), 2) requires reliable methods to optimize the model. Therefore, an improved EM is developed to predict the RUL of rolling bearings. Firstly, an adaptive method based on kurtosis and root mean square (RMS) of bearing vibration signals is used to determine the appropriate FPT. Secondly, gradient descent method is used to reliably optimize the EM. A commonly used bearing degradation datasets are analyzed to show the advantages of the present method. Compared with the traditional EM, the method can not only adaptively determine FPT, but also predict RUL more accurately.” Gang Wang teaches that rolling bearing RUL prediction using an exponential model depends on an appropriate first prediction time. Pg. 3 3.1 FPT selection section of Gang Wang states “In order to use EM to predict the RUL of rolling bearings, FPT must be determined first. The rolling bearings degradation state after FPT is the fault stage, in which the prediction of RUL is implemented. In the normal operation stage of rolling bearings, the bearing signal is irregular vibration caused by environmental noise, and the RMS hardly changes with time. However, in the fault stage, the signal-to-noise ratio (SNR) of the fault signal increases with time, resulting in an exponential rise of RMS with time.” Pg. 5 4. Experimental evaluation section of Gang Wang states “The contents observed on the rolling bearings components before and after the experiment are shown in Fig. 5. During the experiment, the vibration signals of rolling bearings are recorded by acceleration sensor. The acceleration sensor performs sampling once every 10 s, and the sampling frequency is 25.6 kHz. Therefore, each sample contains 2560 data points, that is, 0.1 s. All the rolling bearings signal process and EM construction and RUL prediction are completed by python 3.6 program: https://github.com/famer3riots/PHM-RUL-Prediction-by-EM. In addition, TensorFlow 2.3 is used to construct and train the EM.” Gang Wang describes a normal operation stage before FPT and a fault stage after FPT in which RUL prediction is performed. The normal operation and fault stages correspond respectively to the claimed health and degradation stages. Therefore, the FPT functions as boundary dividing the two stages.)
extracting, from the original state monitoring data of the mechanical component, state monitoring data in the degradation stage of the mechanical component based on the first predicting time, (Pg. 3 of Gang Wang states “In order to use EM to predict the RUL of rolling bearings, FPT must be determined first. The rolling bearings degradation state after FPT is the fault stage, in which the prediction of RUL is implemented. In the normal operation stage of rolling bearings, the bearing signal is irregular vibration caused by environmental noise, and the RMS hardly changes with time. However, in the fault stage, the signal-to-noise ratio (SNR) of the fault signal increases with time, resulting in an exponential rise of RMS with time. In addition, the kurtosis of signals is a very sensitive feature to early fault [7]. As shown in Fig. 6, the kurtosis of vibration signals has changed significantly before the RMS has changed significantly in the early stage of fault. Therefore, observing whether RMS and kurtosis of the signals are within a normal range can determine whether the fault occurs. The rolling bearings must work in the normal operation stage at the beginning of service. The RMS and kurtosis of known vibration signals collected in the past is used to determine the normal range of RMS and kurtosis at the current moment. In the process of rolling bearings operation, due to the interference of environmental noise, sometimes the RMS or kurtosis of the sampled signals can reach the condition of judging fault.” Gang Wang teaches that the observations after FPT belong to the fault or degradation stage and that RUL processing is performed during that stage. While Gang Wang does not expressly use the word “extracting” , selecting the observations occurring after the determined FPT for degradation stage processing would have been the predictable implementation of Gang Wang’s stage based RUL method. )
Ma teaches that
building a hybrid driven prediction model comprising a fully connected layer, … , a multi-head attention mechanism module, a feedforward module and a fully connected regression layer; ([0016] of Ma states “The network structure of the multi-scale hybrid attention mechanism model (the network structure diagram is shown in FIG. 3 a ) comprises a position encoding layer, a feature extraction layer and a regression prediction layer;” [0017] of Ma states “Firstly, mapping the sample X∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
n*k to a higher dimensional space Y∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
n*d through a linear layer so that the data dimension d can be evenly divided by the number H of subsequent attention heads:Y=XWY(3)wherein WY∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
k*d is a trainable projection matrix;” [0019] of Ma states “The feature extraction layer comprises two parts: a multi-head hybrid attention mechanism and a multi-scale convolutional neural network, and residual connection and layer normalization are added simultaneously at the end positions of the two parts to suppress overfitting; the multi-head hybrid attention mechanism is formed by mixing a multi-head self attention mechanism and a multi-head external attention mechanism.” [0030] of Ma states “Firstly, expanding the result MultiScaleConv∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
n*d obtained in step 3.2 as F∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
1*(n*d), and then, calculating the result through a two-layer fully connected neural network to obtain the predicted value RUL of the remaining useful life of an aero-engine: RUL=Relu(FW2+b1)W2+b2(16) wherein W1∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
(n*d)*d 1 is the projection matrix of the first layer of fully connected neural network, b1∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
1*d 1 is the bias of the first layer of fully connected neural network, W2∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
d 1 *1 is the projection matrix of the second layer of fully connected neural network, b2∈
PNG
media_image1.png
38
29
media_image1.png
Greyscale
1*1 is the bias of the second layer of fully connected neural network, the projection matrixes and the biases are both trainable, and Relu is an activation function;” The first fully connected stage performs feedforward feature transformation while the second fully connected stage maps the transformed features to the RUL regression output. Under BRI, the first stage corresponds to the feedforward module and the second stage corresponds to the fully connected regression layer. )
training and testing the hybrid driven prediction model by using the neural network training data set, to obtain a trained hybrid driven prediction model; ([0004] of Ma states “In this method, samples are constructed by sliding time windows on the historical sensor data of aero-engines, and then features are extracted by convolutional neural network. Finally, the remaining useful life is predicted through the full connection layer. Convolutional neural network is a kind of feedforward neural network through convolution calculation, which is inspired by the mechanism of receptive field in biology. It has translation invariance, uses convolution kernel, makes maximum use of local information, and retains plane structure information” [0016] of Ma states ”The network structure of the multi-scale hybrid attention mechanism model (the network structure diagram is shown in FIG. 3 a ) comprises a position encoding layer, a feature extraction layer and a regression prediction layer;” [0019] of Ma states “The feature extraction layer comprises two parts: a multi-head hybrid attention mechanism and a multi-scale convolutional neural network, and residual connection and layer normalization are added simultaneously at the end positions of the two parts to suppress overfitting; the multi-head hybrid attention mechanism is formed by mixing a multi-head self attention mechanism and a multi-head external attention mechanism.” [0030] of Ma states “(3.3) Regression Prediction Layer Firstly, expanding the result MultiScaleConv∈Rn*d obtained in step 3.2 as F∈R1*(n*d), and then, calculating the result through a two-layer fully connected neural network to obtain the predicted value RUL of the remaining useful life of an aero-engine:” [0032] of Ma states “Firstly, inputting the samples obtained in step 1.3 into the multi-scale hybrid attention mechanism model constructed in step 1.3 in batches to obtain the predicted value RUL, then calculating the MSE loss value, and conducting gradient update of the model using an adaptive moment estimation optimizer to complete an iterative training session; and setting the total number of model training iterations, and iteratively training the model several times;” [0033] of Ma states “At the on-line testing stage, calculating the output value by preprocessing the data in step 1 and inputting the data into the multi-scale hybrid attention mechanism model trained in step 4 according to the real-time data collected by the aero-engine sensor, wherein the output value is the predicted value of the remaining useful life of the aero-engine.” Pg. 7 of Wang states “Generally, the rolling bearing has a performance degradation failure and burst failure mode, the invention combines with the characteristics bearing vibration signals and deep learning and LSTM each advantage, digging deep characteristic and constructing trend quantitative health index, paintable bearing degradation trend and realizing the prediction of the rolling bearing RUL. FIG. 3 is a block diagram of the whole process frame. The specific process steps are as follows: 1) selecting the part data of each working condition of the rolling bearing vibration signals the training set, and performing FFT conversion to the original vibration signals of the training set to obtain the frequency domain amplitude value signal.” Wang and Ma teach training and testing the combined neural network model. )
and inputting the processed new state monitoring data into the trained hybrid driven prediction model to predict the remaining useful life of the mechanical component (Pg. 3 of Wang states “5) performing FFT conversion to the non-full-life time domain vibration signals of different working conditions in the test set, obtaining the frequency domain amplitude signal, and performing normalization processing, step 3) digging the deep characteristic, combining the trend quantitative health index model of step 4), obtaining the trend quantitative health index of the test set. using the moving average method (average, MA) to smooth [17], reducing oscillation to improve the remaining life prediction precision. and 6) fitting the performance degradation trend the rolling bearing by the polynomial curve [18] to predict the RUL of the rolling bearing.” [0033] of Ma states “At the on-line testing stage, calculating the output value by preprocessing the data in step 1 and inputting the data into the multi-scale hybrid attention mechanism model trained in step 4 according to the real-time data collected by the aero-engine sensor, wherein the output value is the predicted value of the remaining useful life of the aero-engine.” It would have been obvious to input both the updated EKF state estimate and FFT derived features into the trained model. Generating the sample categories of features during training and prediction would allow newly acquired data to be presented in the feature representation on which the model was trained.)
It would have been obvious to one with ordinary skill in the art before the effective filing date of the invention to combine the teachings of Singleton, Ma, Gang Wang, and Wang because all references are directed to RUL prediction for mechanical assets. Singleton teaches an EKF based degradation modeling and state estimation framework for RUL prediction with equations. Wang teaches obtaining frequency domain information from monitored bearing signals using FFT and using those signals as input for neural network-based RUL prediction. Gang Wang teaches determining first prediction time as a key for implementing RUL prediction model. Ma teaches constructing training samples from historical sensor data and training RUL prediction model including attention based and fully connected prediction components. One with ordinary skill in the art would be motivated to incorporate the teachings of Singleton, Ma, Gang Wang, and Wang because they solve different but complementary parts of the same RUL prediction pipeline. Combining them would have predictably improved the overall pipeline by using degradation stage data that are more relevant, extracting signal features that better reflect degradation, and training hybrid models on those more useful inputs to produce accurate RUL predictions.
Regarding claim 2, the rejection of claim 1 is incorporated herein. Furthermore, the combination of Singleton, Wang, Gang Wang, and Ma teaches
wherein modeling the degradation process of the mechanical component by fitting the variance feature sequence extracted from the original state monitoring data through the exponential random model, to obtain the degradation model of the mechanical component, and establishing the system state space equation based on the the degradation model of the mechanical component comprises: using the exponential random model
PNG
media_image2.png
32
57
media_image2.png
Greyscale
to model the degradation process of the mechanical component,
PNG
media_image3.png
32
63
media_image3.png
Greyscale
being parameters related to a health state of the mechanical component in the degradation process; and building the system state space equation
PNG
media_image4.png
85
258
media_image4.png
Greyscale
based on the exponential random model
PNG
media_image2.png
32
57
media_image2.png
Greyscale
, a state vector at a moment k being
PNG
media_image5.png
43
231
media_image5.png
Greyscale
f and h being nonlinear functions,
PNG
media_image6.png
33
45
media_image6.png
Greyscale
being a state vector at a moment k-1,
PNG
media_image7.png
29
45
media_image7.png
Greyscale
being a system input at the moment k-1,
PNG
media_image8.png
32
57
media_image8.png
Greyscale
being a random zero mean error at the moment k-1,
PNG
media_image9.png
27
33
media_image9.png
Greyscale
being a measured value at the moment k, and
PNG
media_image10.png
23
26
media_image10.png
Greyscale
being a measurement error at the moment k. (Pg. 1785 C.EKF Parameter Learning section of Singleton states “For the two different types of features, i.e., vibration and entropy, time-dependent degradation models are obtained through curve fitting. For the variance feature, an exponential of the form aebt was found to be the most suitable, whereas for the TF entropy feature, a curve in the form of a − be−ct was more suitable… The parameters of the degradation function are updated with each new measurement. To accomplish this, a state vector x containing the equation for the curve fit as well as the unknown parameters describing this degradation model at each time point are defined for each feature. For the variance feature, the parameters ak and bk of the exponential curve are used to define the state vector as [43], [53] xk = [akebkk ak bk]T (5) and for the entropy feature, we have xk = [ak – bke−ckk ak bk ck]T (6) both with the measurement equation given by zk = h(xk) = xk(1). (7)” It is also noted that there is no input to this system; hence, in our case, uk defined in (3) is equal to zero. With each time step, the parameters of the degradation model are updated to form a new model, i.e., fk, and an estimate of the next state, i.e., ˆxk, is calculated. Functions f and h are then locally linearized about that estimate to produce Fk and Hk” Pg. 1783 D. Extended Kalman Filtering section of Singleton states “The EKF equation in the presence of process noise and measurement noise is xk = f(xk−1, uk−1) +wk−1 (3) where xk is the state being estimated, f is a nonlinear function of states, uk is the input at time sample k, wk is random zero mean noise with covariance matrix Qk. In EKF, the relationship between system states (xk) and measurements (zk) can also be nonlinear, i.e., zk = h(xk)+vk (4) where zk is the measurement, h is a measurement function that is a nonlinear function of states, and vk is a zero-mean random process described by the measurement noise covariance matrix”)
Regarding claim 3, the rejection of claim 2 is incorporated herein. Furthermore, the combination of Singleton, Wang, Gang Wang, and Ma teaches
locally linearizing, based on the system state space equation, the nonlinear functions
PNG
media_image11.png
31
106
media_image11.png
Greyscale
at the moment k about a state prior estimate
PNG
media_image12.png
37
32
media_image12.png
Greyscale
, to obtain corresponding Jacobian metrices
PNG
media_image13.png
34
101
media_image13.png
Greyscale
; building a prediction and update equation of an extended Kalman filter according to the Jacobian matrices
PNG
media_image13.png
34
101
media_image13.png
Greyscale
; (Pg. 1785 C.EKF Parameter Learning section of Singleton states “With each time step, the parameters of the degradation model are updated to form a new model, i.e., fk, and an estimate of the next state, i.e., ˆxk, is calculated. Functions f and h are then locally linearized about that estimate to produce Fk and Hk by
PNG
media_image14.png
115
329
media_image14.png
Greyscale
.” Pg. 1783 D. Extended Kalman Filtering section of Singleton states “To carry out the normal KF operations, the nonlinear functions, i.e., f and h, must be locally linearized around the estimated state by calculating their respective Jacobian, producing matrices F and H, respectively. One important point in the implementation of EKF is the choice of the initial parameters, as the speed of convergence depends on the initial estimate ˆx0 and the uncertainty matrix P0 [52].”)
and alternately executing, based on the prediction and update equation, a prediction and update process of the extended Kalman filter to continuously update a predicted state vector, so as to obtain the optimal state estimate. (Pg. 1785 C. EKF Parameter Learning section of Singleton states “The parameters of the degradation function are updated with each new measurement. To accomplish this, a state vector x containing the equation for the curve fit as well as the unknown parameters describing this degradation model at each time point are defined for each feature. For the variance feature, the parameters ak and bk of the exponential curve are used to define the state vector” Pg. 1785-86 D. RUL Prediction section of Singleton states “An overall view of the algorithm and all its steps are given below: 1) Initialize x0 and P0. 2) Predict the next state, ˆxk, and uncertainty matrix, Mk, i.e., ˆxk =f(ˆxk−1) + wk−1 (14) Mk =Fk−1Pk−1Fk−1 + Qk−1. (15) 3) Take in measurement zk. 4) Update the predictions and their uncertainties using the Kalman gain, Kk, i.e., Kk =MkHTkHkMkHTk+ Rk−1 (16) ˆxk ←− ˆxk +Kk(zk −Hkˆxk) (17) Pk =(1 −KkHk)Mk. (18) 5) The current value of the feature state is extrapolated out to failure threshold. The total number of time steps, i.e., n, required to reach the failure threshold is taken as the RUL at time k, i.e., ¯γo = ˆxk+n = fk(ˆxk+n−1) + wk+n. (19) 6) Calculate the confidence intervals of RUL predictions (see Section III-E). 7) Repeat process starting at step 3.” Singleton teaches alternatively executing a prediction and update process of the extended Kalman filter to continuously update a predicted state vector and update state estimate. )
Regarding claim 5, the rejection of claim 1 is incorporated herein. Furthermore, the combination of Singleton, Wang, Gang Wang, and Ma teaches
constructing the neural network training data set
PNG
media_image15.png
76
441
media_image15.png
Greyscale
of all the mechanical components according to the optimal state estimate
PNG
media_image16.png
43
112
media_image16.png
Greyscale
and the frequency domain data
PNG
media_image17.png
44
108
media_image17.png
Greyscale
in the degradation stage of the mechanical component, a moment
PNG
media_image18.png
31
23
media_image18.png
Greyscale
being the first predicting time,
PNG
media_image19.png
27
27
media_image19.png
Greyscale
being a length of a variance feature sequence, Q being a number of the mechanical components, and
PNG
media_image20.png
36
38
media_image20.png
Greyscale
being the remaining useful life of the mechanical component i at the moment k. (Wang states “
PNG
media_image21.png
57
406
media_image21.png
Greyscale
1) selecting part data of rolling bearing vibration signals different working conditions as the training set, and performing FFT conversion on the original vibration signals of the training set to obtain the frequency domain amplitude signal; 2) the frequency domain amplitude signal after performing normalization processing as characteristic input, life percentage as output training model, formula wherein xt belongs to RN * 1 represents the N-dimensional feature input of a bearing at time t, N=2048, yt belongs to [0, 1] represents the bearing at time t of life degradation percentage output; Dtra vibration signals the data of a bearing of a certain working condition in training set, R is frequency domain amplitude characteristic matrix; T is the running time of the bearing full service life;” Wang shows the constructing the neural network training dataset and obtaining the frequency domain data in the degradation stage. Pg. 1785 C.EKF Parameter Learning section of Singleton states “For the two different types of features, i.e., vibration and entropy, time-dependent degradation models are obtained through curve fitting. For the variance feature, an exponential of the form aebt was found to be the most suitable, whereas for the TF entropy feature, a curve in the form of a − be−ct was more suitable… With each time step, the parameters of the degradation model are updated to form a new model, i.e., fk, and an estimate of the next state, i.e., ˆxk, is calculated.” Singletone teaches generating and updating the EKF state estimate which corresponds to the optimal state estimate for each time given. Singletone also teaches a variance feature tracked over time as part of the degradation feature sequence, which length of it could be denoted as ni . Pg. 3 3.1 FPT selection section of Gang Wang states “In order to use EM to predict the RUL of rolling bearings, FPT must be determined first. The rolling bearings degradation state after FPT is the fault stage, in which the prediction of RUL is implemented.” Gang Wang directly states that FPT should be determined so it could be accounted for constructing data set. The notation I = 1,…,Q merely indexes multiple mechanical components and would have been obvious when constructing a training dataset for multiple mechanical components. Regarding y being the remaining useful life of the mechanical component at the moment k, Pg. 1786 of Singletone algorithm states “The current value of the feature state is extrapolated out to failure threshold. The total number of time steps, i.e., n, required to reach the failure threshold is taken as the RUL at time k, i.e.,” It gives an explicit RUL estimate at time K before meeting the failure threshold.)
Claim 7 recites substantially similar subject matter to claim 1 respectively and is rejected with the same rationale, mutatis mutandis.
Claims 9,10, 12 recite substantially similar subject matter to claim 2,3, 5 respectively and are rejected with the same rationale, mutatis mutandis.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 BYUNGKWON HAN whose telephone number is (571)272-5294. The examiner can normally be reached M-F: 9:00AM-6PM PST.
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, Li B Zhen can be reached at (571)272-3768. 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.
/BYUNGKWON HAN/ Examiner, Art Unit 2121
/Li B. Zhen/ Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights