DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . Claims 1-17 ae pending in this Application. Claim 5 has been withdrawn from consideration.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Restriction Requirement
Applicant’s election without traverse of claims 1-4 and 6-17 in the reply filed on 02/17/2026 is acknowledged.
The claim restrictions were reconsidered and the restriction is deemed proper and is therefore made final. Claim 5 has been withdrawn from consideration.
Information Disclosure Statement
The information disclosure statement filed on 10/19/2022 fails to comply with the provisions of 37 CFR 1.97, 1.98 and MPEP § 609 because the reference Kircher et al “Detecting process instabilities in industrial gas metal arc welding timer series” was not properly received. The received document is blank. It has been placed in the application file, but the information referred to therein has not been considered as to the merits. Also, the reference EP 3736645 mentioned in the disclosure and cited in the same IDS, was not received and does not comply with the provisions previously mentioned above. It has been placed in the application file, but the information referred to therein has not been considered as to the merits.
Applicant is advised that the date of any re-submission of any item of information contained in this information disclosure statement or the submission of any missing element(s) will be the date of submission for purposes of determining compliance with the requirements based on the time of filing the statement, including all certification requirements for statements under 37 CFR 1.97(e). See MPEP § 609.05(a).
Claim Objections
Claim 5 while being non-elected is objected to because of the following informalities: objected under 37 CFR 1.75(c) as being in improper form because a dependent claim should be referring back to a preceding claim. See MPEP § 608.01(n). In this case, Claims 5 does not refer back to a preceding claim. However, the text and limitations recited in claim 5 seems to be referring to claim 4. Thus, this claim will be interpreted as being dependent on claim 4.
Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-4 and 6-17 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract without significantly more.
Claims 1, 15, 16, and 17 recite:
“ determining a reconstruction error based on the at least one process signal data set and the at least one reconstructed process signal data set; and
detecting an anomaly of the laser machining process based on the determined reconstruction error”,
Under the broadest reasonable interpretation, the terms of the claims are presumed to have their plain meaning consistent with the specification as it would be interpreted by one of ordinary skill in the art. See MPEP 2111. Under the broadest reasonable interpretation, these limitations, as drafted, are a process that, under its broadest reasonable interpretation (BRI), covers mathematical relationships which is identified as an example of mathematical concepts grouping of abstract ideas and/or steps that are easily performed in the mind and thus covers a mental process. See MPEP 210604(a)(2), III. For instance, steps a) and b) are directed to calculation of an error value between two values (estimated and actual) and to determination of an anomaly based on comparing the calculated value with a threshold. These steps are exemplified as mathematical values and mathematical expression or calculation which are easily calculated and detecting the anomaly comprises the comparing of a value with a threshold according to the disclosure (see [003] and [0046]). Also, these steps can be performed by a user mentally by looking at the output of the signals compared to an original signal as shown in Fig. 4 and can easily determine that there is anomaly in the process or by using pen and paper to determine a reconstruction error. Therefore, As discussed above, the broadest reasonable interpretation of steps a) and b) is that this step fall within the mental process groupings of abstract ideas because they cover concepts performed in the human mind, including observation, evaluation, judgment, and opinion and/or mental steps with the use of pen and paper (See MPEP 2106.04(a)(2), subsection III).
This judicial exception is not integrated into a practical application because the additional elements such as “inputting at least one process signal data set of the laser machining process into an autoencoder formed by a deep neural network and generating a reconstructed process signal data set by means of said autoencoder” which are recited in high level of generality and do not impose any meaningful limits on practicing the abstract idea. The claim does not provide any details about how the trained ANN operates or how the reconstruction is made, and inputting data to an encoder and performing the reconstruction which is the output of the ANN are well understood, routine and conventional activities (see 2106.05(d)). Claims 1, 15, and 16-17 further includes the additional element of machining a workpiece with a laser beam, detecting anomalies of the laser process, including a sensor to sense process signals of the laser machining process, recited at high level of generality, represents mere instructions to “apply” the abstract idea or to generally link the use of the judicial exception to a particular field of use and/or technological environment such as a laser machine and its components and cannot provide an inventive concept as stated by the courts (see MPEP 2106.05(f) and 2106.05(h)). Claim 16 recites the additional element of at least one processor configured to carry out the method for monitoring the laser machining process according to claim, recited at high level of generality and represents no more than instructions “to apply” the abstract idea on a computer or to generally link the use of the judicial exception to the technological environment of a computer cannot provide an inventive concept as stated by the courts (see MPEP 2106.05(f) and 2106.05(h)).
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements such as “inputting at least one process signal data set of the laser machining process into an autoencoder formed by a deep neural network and generating a reconstructed process signal data set by means of said autoencoder” which are recited in high level of generality and do not impose any meaningful limits on practicing the abstract idea. The claim does not provide any details about how the trained ANN operates or how the reconstruction is made, and inputting data to an encoder and performing the reconstruction which is the output of the ANN are well understood, routine and conventional activities (see 2106.05(d)). For instance, autoencoder performing data reconstruction, generating a reconstructed signals using the autoencoder are steps that are re well understood, routine and conventional activities. For instance, Zhang et al, "A Low-Cost Welding Status Monitoring Framework for High-Power Disk Laser Welding (December 2018),” teaches a system and method for detecting anomalies for a laser welding process comprising an autoencoder (see Fig. 1 and 2) and a classifier (fig. 3) and for classifying the anomalies and inputting at least one process signal data set of a process into an autoencoder formed by a deep neural network, generating a reconstructed process signal data set by means of said autoencoder. Zhang et al, "Real-time monitoring of high-power disk laser welding statuses based on deep learning framework", 2020”, teaches an autoencoder, inputting at least one process signal data set of the laser machining process into an autoencoder formed by a deep neural network, generating a reconstructed process signal data set by means of said autoencoder (see Fig. 2), to detect anomalies combined with a classifier to classify anomalies (fig. 9) in laser welding system process (page 807-808). Allahdadian et al (US 20220156578) teaches inputting at least one process signal data set of a process into an autoencoder formed by a deep neural network, generating a reconstructed process signal data set by means of said autoencoder (see 0032). Gregson et al (US 20210342570) teaches an autoencoder combined with a classifier (see fig. 5 and 0035-0036), and inputting at least one process signal data set of a process into an autoencoder formed by a deep neural network, generating a reconstructed process signal data set by means of said autoencoder (see 034). Jun et al (US 20210256991) teaches an autoencoder combined with a classifier to detect anomalies (see fig. 8 and 0057, 0060, and 0065), and i inputting at least one process signal data set of a process into an autoencoder formed by a deep neural network, generating a reconstructed process signal data set by means of said autoencoder (see 0054). Fan et al (US 20210049452) teaches an autoencoder combined with a classifier to detect anomalies (see fig. 2A and 0023, 0032), and inputting at least one process signal data set of a process into an autoencoder formed by a deep neural network, generating a reconstructed process signal data set by means of said autoencoder. Horster et al (US 20100299303) teaches inputting at least one process signal data set of a process into an autoencoder formed by a deep neural network, generating a reconstructed process signal data set by means of said autoencoder (see [0057] and [0064]). Claims 1, 15, and 16-17 further includes the additional element of machining a workpiece with a laser beam, detecting anomalies of the laser process, including a sensor to sense process signals of the laser machining process, recited at high level of generality, represents mere instructions to “apply” the abstract idea or to generally link the use of the judicial exception to a particular field of use and/or technological environment such as a laser machine and its components and cannot provide an inventive concept as stated by the courts (see MPEP 2106.05(f) and 2106.05(h)). Claim 16 recites the additional element of at least one processor configured to carry out the method for monitoring the laser machining process according to claim, recited at high level of generality and represents no more than instructions “to apply” the abstract idea on a computer or to generally link the use of the judicial exception to the technological environment of a computer cannot provide an inventive concept as stated by the courts (see MPEP 2106.05(f) and 2106.05(h)). Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claims 2-4 and 6-14, respectively and thus recites the limitations and the abstract ideas of their respective parent claims.
Claim 2 recites the additional element of “measuring at least some of the process signals of the process signal data set, and/or transmitting at least some of the process signals of the process signal data set from a control” recited at high level of generality and represents mere data gathering and mere data outputting and thus are insignificant extra-solution activity that do not impose any meaningful limitations on the claim (. See MPEP 2106.05(g) “whether the limitation is significant’). Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claim 3-4 recites the additional elements of “determining a quality feature of the laser machining process, and evaluating the quality feature as valid when no anomaly is detected in the step of detecting an anomaly of the laser machining process, and evaluating the determined quality feature as not valid when an anomaly of the laser machining process is detected in the step of detecting an anomaly, wherein: the step of determining a quality feature of the laser machining process is by means of a regressor formed by a neural network, and a value for the quality feature is determined, and/or the step of determining a quality feature of the laser machining process is by means of a classifier formed by a neural network, and a classification value for the quality feature is determined”, these limitations, as drafted, are a process or steps that, under its broadest reasonable interpretation, covers mathematical relationships which are part of the group of mathematical concepts abstract ideas or mental process and/or steps that are easily performed in the mind and thus covers a mental process. For instance, this step involves the step of calculating a quality value, and evaluating the value comprises a comparison of values to make a determination of valid/pass or invalid/no pass. These additional elements also can be performed mentally by observing a difference in the data 9collection of data) and making a judgement when the data input and the reconstructed data are different. The additional element of using a regressor or a classifier to determining a quality feature recited at a high level of generality, wherein t The claim does not provide any details about how the trained ANN operates or how the detection/determination is made, and the plain meaning of “detecting” encompasses mental observations or evaluations, e.g., a computer programmer’s mental identification of a feature in a data set. Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claims 6-7 further recites the additional elements of “wherein: said autoencoder and at least one of said regressor and said classifier are parallel to each other and/or said regressor and/or said classifier determines the quality feature based on the at least one process signal data set and/or said autoencoder and at least one of said regressor and said classifier have a common input layer” recited at high level of generality, wherein said autoencoder and at least one of said regressor and said classifier are trained with the same data”, The claim does not provide any details about how the trained ANN (regressor or classifier) operates or how the determination of the quality feature is made, and the plain meaning of “determining” encompasses mental observations or evaluations, e.g., a computer programmer’s mental identification of a quality feature based on a data set. Training the encoder and at least one of said regressor and said classifier are trained with the same data is a tangential limitation which is recited at high level of generality which does not impose any boundaries or limits to the abstract idea of claim 1 . Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claims 8-9 further recites “wherein the step of determining a reconstruction error comprises…and/or determining a mean absolute or squared deviation of the at least one process signal data set from the at least one reconstructed process signal data set, and/or determining a signed, absolute or squared deviation summed up along the time axis, and/or determining a Mahalanobis distance” and wherein the step of determining a reconstruction error comprises determining a Mahalanobis distance with respect to: a deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or individual characteristic values of the reconstruction error; and/or encoding of a process signal data set”, covers mathematical relationships which is identified as an example of mathematical concepts grouping of abstract ideas and/or steps that are easily performed in the mind and thus covers a mental process. See MPEP 210604(a)(2), III. Determining a mahalanobis distance comprises the calculation and use of the mahalanobis formula. Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claim 10 further recites the additional element of “wherein said method comprises determining parameters mean vector and covariance matrix of the Mahalanobis distance using defect-free or labeled data sets”, these limitations, as drafted, are a process that, under its broadest reasonable interpretation (BRI), covers mathematical relationships which is identified as an example of mathematical concepts grouping of abstract ideas. Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claim 11, further recites the additional element of “wherein the reconstruction error for individual dimensions is determined separately and/or based on a metric and/or by means of a fast Fourier transformation and/or a wavelet transformation”, these limitations, as drafted, are a process that, under its broadest reasonable interpretation (BRI), covers mathematical relationships which is identified as an example of mathematical concepts grouping of abstract ideas. The Fast Fourier Transform (FFT) is an optimized algorithm used to compute the Discrete Fourier Transform (DFT). Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claims 12 recites the additional element of normalizing the reconstruction error with respect to the process signal data set, and/or the step of determining a reconstruction error comprises: filtering at least part of the process signal data set and/or the reconstructed process signal data set, and based thereon, determining the reconstruction error”, recited at high level of generality and these limitations, as drafted, are a process that, under its broadest reasonable interpretation (BRI), covers mathematical relationships which is identified as an example of mathematical concepts grouping of abstract ideas. Normalization involves the calculation and scaling of data values. Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
Claims 13-14 further recites “determining a degree of abnormality; wherein detecting an anomaly of the laser machining process is based on the determined degree of abnormality, wherein the determining a degree of abnormality is based on a weighted summation or on a Mahalanobis distance with respect to individual characteristic values for the reconstruction error”, recited at high level of generality and these limitations, as drafted, are a process that, under its broadest reasonable interpretation (BRI), covers mathematical relationships which is identified as an example of mathematical concepts grouping of abstract ideas. Calculating a degree of abnormality/value using the mahalanobis distance . Determining a mahalanobis distance comprises the calculation and use of the mahalanobis formula. Also, determining the anomaly based on the calculated degree of abnormality is a step of comparing a value such as degree of abnormality against a threshold metric to decide if an anomaly exists. Accordingly, these additional elements do not integrate the abstract idea into a practical application, do not amount to significantly more than the judicial exception, and do not impose any meaningful limits on practicing the abstract idea. Therefore, the claims are not patent eligible.
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.
Claim(s) 1-3, 8, 11 and 15-17 are rejected under 35 U.S.C. 103 as being unpatentable over Hsu (US 20170032281) in view of Kirchner et al (“Detecting process instabilities in industrial gas metal arc welding time series”, 2020).
As per claim 1, Hsu teaches a method for monitoring a laser machining process, said method comprising the steps of (see Fi.3 200 and see [0069] “The welding system 200 of FIG. 2 may be configured to cut material (e.g., as a plasma cutter) or form a solder or brazed joint, form a weld (e.g., weld joint 212) between two components in a weldment by any known welding technique including,… laser welding…”):
inputting at least one process signal data set of the laser machining process into an autoencoder formed by a deep neural network (see Fig. 4b deep neural network receives X data and y data; also, see Figs. 4c-4e data is ingested in a DNN; also, see [0075] “… The welding equipment 210 may be configured to communicate the welding data to the analytics computing platform 234 via the communication network 232 for processing, while still preserving traceability to the weldments…” and see [0079] “With reference to FIGS. 4a through 4d, example welding process systems are illustrated where the welding cell 406 records the raw welding process signals and data from the sensors in the welding operation, and transmit the features (x) along with tag data (or weld process metadata) to a central location, e.g., analytics computing platform 234 via a communication network 232…”; also, see[0084-0087] “ It is possible for example to predict the weld quality benefit or welding equipment service benefit of a Regulated Metal Deposition (RMD) or Controlled Short Circuit (CSC) welding process from one plant of a fabricator to another plant using similar welding process and requirements. In addition, application servers or app servers may provide Matlab or Octave application services or shell scripts (e.g. supervised learning such as LVQ, NARX and RNN; unsupervised learning such as self-organizing maps and competitive layers; and deep learning such as convolutional neural net and auto encoders) to Matlab or Octave operators located remotely at the labor center 602, which may include one or more of machine learning analysts,…”; also, see [0057-0058] “…unsupervised machine learning algorithms process unlabeled training data, typically for data pre-processing such as data compression, integration, feature identification, cataloging, transformation or for data clustering and data mining (e.g., Principle Component Analysis or PCA, clustering such as clustering by k-means, hierarchical, conceptual, probability-based and Bayesian). One example of data clustering analysis is to identify batch-to-batch differences in manufacturing. Another example is anomaly detection in weld quality or equipment condition or usage or imminent failure; or welding consumables such as gas, wire, flux, tip, nozzle and liner; or parts being welded; or weld fixture being used. Yet another example is the grouping of trained/skilled/certified/high performing weld operators from novice/underperforming/un-certified weld operators. Yet another example is the detection of any deviation from weld procedure specification (WPS)…”; [0060] “Weld production knowledge machine learning algorithms may be used to predict and/or identify predetermined characteristic of said welding equipment or welding personnel, such as, inter alia, tool life, weld quality (e.g., passing or failing the WPS or compliance with production specifications), weldment quality, weld tool consumable life, welding equipment (and its component), service condition/interval, welding equipment reliability such as MTTR/MTBF, decisions and actions by weld personnel...”; also, see [0064] “…Accordingly, an unsupervised learning algorithm may be used to detect anomalies in welding operations, such as those in weld quality, in arc time pattern (active welding)…”);
and detecting an anomaly of the laser machining process (see [0059]-[0060] and [0064-0065] “…Accordingly, an unsupervised learning algorithm may be used to detect anomalies in welding operations, such as those in weld quality, in arc time pattern (active welding)…”; also, see [0088] “…, weld quality prediction, anomaly detection, and activity tracking in a manufacturing and production environment…”; also, see [0092]).
Hsu clearly teaches that the analytics for predicting anomalies in the laser process can use an auto-encoder algorithms, wherein an autoencoder is a very well-known unsupervised ANN that performs a reconstruction process and determined a reconstruction error conventionally used to detect abnormal signals or anomalies, However, Hsu does not explicitly teach generating a reconstructed process signal data set by means of said autoencoder, determining a reconstruction error based on the at least one process signal data set and the at least one reconstructed process signal data set and detecting an anomaly of the laser machining process based on the determined reconstruction error.
However, Kirchner teaches a method and system to detect instabilities in a machine process comprising an autoencoder, generating a reconstructed process signal data set by means of said autoencoder (see the Abstract “… The latent space of the autoencoder and reconstruction of the time series are used to detect process instabilities…”; also, see page 152 Col 1-2 “…3.1. Autoencoder architecture… The second component is the decoder, which recon-structs the input by using the function k = g(h)…. An encoder learns a vector representation that is used by the decoder to reconstruct the input time series…), determining a reconstruction error based on the at least one process signal data set and the at least one reconstructed process signal data set (see page 152 Col 2 “..series. The reconstruction error is deployed for determining the quality of the architecture… 3.2. Anomaly detection Processing data through the autoencoder architecture leads to a reconstructed time series. This reconstruction can be directly compared to the input data, and an anomaly score can be calculated that compares the reconstruction with the original input. The idea of training the autoencoder with only normal input data leads to a failure of reconstructing anomalous inputs reasonably.”), and detecting an anomaly of the machining process based on the determined reconstruction error (see page 152 Col 2 “..series. The reconstruction error is deployed for determining the quality of the architecture… 3.2. Anomaly detection Processing data through the autoencoder architecture leads to a reconstructed time series. This reconstruction can be directly compared to the input data, and an anomaly score can be calculated that compares the reconstruction with the original input. The idea of training the autoencoder with only normal input data leads to a failure of reconstructing anomalous inputs reasonably…”; also, see Fig. 2 and see page 153 Col 1 “ The reconstruction error e of the test measurements calculated by using the formula… determines a value to describe the reconstruction at each specific point of a recording. By using a threshold tr, every measurement that exceeds the threshold is marked as an anomaly (e j > tr). The main advantage of the reconstruction model is that it is easily possible to determine the exact position of the anomalous pattern within a time series….; also, see page 154 “…The sum of the instabilities represents the total number of detected anomalies exceeding the absolute threshold.).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu’s invention to include an autoencoder, generating a reconstructed process signal data set by means of said autoencoder, determining a reconstruction error based on the at least one process signal data set and the at least one reconstructed process signal data set and detecting an anomaly of the laser machining process of Hsu based on the determined reconstruction error as taught by Kirchner in order to determine anomalies in the laser process of Hsu using an autoencoder since autoencoder perform dimensionality reduction and perform processes in less computation time compared to other algorithms (see page 152 Col 2 “to gain a meaningful dimensionality reduction resulting in more appropriate results than a simple principal component analysis. Especially when recording high-frequency data, which result in a large amount of data to be processed per measurement, a dimensionality reduction within the analysis procedure is useful. In general, machine learning models that have to process smaller spaces consume less memory and computation time”).
As per claim 2, Hsu-Kirchner teaches the method according to claim 1, further comprising: measuring at least some of the process signals of the process signal data set (see [0079] “With reference to FIGS. 4a through 4d, example welding process systems are illustrated where the welding cell 406 records the raw welding process signals and data from the sensors in the welding operation, and transmit the features (x) along with tag data (or weld process metadata) to a central location, e.g., analytics computing platform 234 via a communication network 232.”; also, see [0117]);
and/or transmitting at least some of the process signals of the process signal data set from a control (see [0072] “…In some aspects, certain of the parameters (e.g., weld programs, set points, set up conditions, etc.) and fault or event codes may be transmitted from the robot and/or welding equipment 210 to the analytics computing platform 234 as input features or automatically detected/sensed, thereby obviating the need for welding personnel to manually indicate at least those parameters…”).
As per claim 3, Hsu-Kirchner teaches the method according to claim 1, further comprising the steps:
determining a quality feature of the laser machining process (see [0060] “Weld production knowledge machine learning algorithms may be used to predict and/or identify predetermined characteristic of said welding equipment or welding personnel, such as, inter alia, tool life, weld quality (e.g., passing or failing the WPS or compliance with production specifications”);
and evaluating the quality feature as valid when no anomaly is detected in the step of detecting an anomaly of the laser machining process (see [0060] “weld quality (e.g., passing or failing the WPS or compliance with production specifications”, e.g. passing; see [0074] );
and evaluating the determined quality feature as not valid when an anomaly of the laser machining process is detected in the step of detecting an anomaly (see [0060] “weld quality (e.g., passing or failing the WPS or compliance with production specifications”, e.g. failing; see [0064] “… Accordingly, an unsupervised learning algorithm may be used to detect anomalies in welding operations, such as those in weld quality,…”; [0073] “…The quality classifications may comprise one or more of, for example, a missing weld, misaligned weld, weld is too large or too small (i.e., deviates from a predetermined acceptable range), porosity, undercut, burn thru, lack of fusion, failed bend test, failed tensile test, failed fatigue test, failed Charpy test, distortion out of spec, work piece discoloration, etc.”; also, see [0074] “whether a weld is deemed by to be “acceptable” or “unacceptable” (whether manually inputted, or automatically determined from quality assurance equipment measurements) and label specific pass/fail result per quality inspection criteria”; also, see [0079] “…a quality assurance device 410 may classify the weldment as passing (or failing) in one or more aspects of routine tests and communicate the test results (y1, y2) along with tag data (or weld quality metadata) to the analytics computing platform 234….”; also, see [0087] “… to provide services such weld quality prediction, maintenance prediction and data mining (for unexpected anomaly detection and alarm). … One example is to run supervised learning algorithm to train hypothesis h.sub.θ to predict weld quality shown in Equation 2…”; also, see [0093]).
As per claim 8, Hsu-Kirchner teaches the method according to claim 1, Kirchner further teaches wherein the step of determining a reconstruction error comprises:
determining a deviation of the at least one process signal data set from the at least one reconstructed process signal data set (see page 153 Col 1 “The reconstruction error e of the test measurements calculated by using the formula ej= xj−x’j; see page 152 Col 2 “..series. The reconstruction error is deployed for determining the quality of the architecture… 3.2. Anomaly detection Processing data through the autoencoder architecture leads to a reconstructed time series. This reconstruction can be directly compared to the input data, and an anomaly score can be calculated that compares the reconstruction with the original input. The idea of training the autoencoder with only normal input data leads to a failure of reconstructing anomalous inputs reasonably.”); and/or determining a mean absolute or squared deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or determining a signed, absolute or squared deviation summed up along the time axis; and/or determining a Mahalanobis distance.
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu’s invention to include wherein the step of determining a reconstruction error comprises: determining a deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or determining a mean absolute or squared deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or determining a signed, absolute or squared deviation summed up along the time axis; and/or determining a Mahalanobis distance as taught by Kirchner in order to determine anomalies in the laser process of Hsu using an autoencoder since autoencoder perform dimensionality reduction and perform processes in less computation time compared to other algorithms (see page 152 Col 2 “to gain a meaningful dimensionality reduction resulting in more appropriate results than a simple principal component analysis. Especially when recording high-frequency data, which result in a large amount of data to be processed per measurement, a dimensionality reduction within the analysis procedure is useful. In general, machine learning models that have to process smaller spaces consume less memory and computation time”) and since the reconstruction error is a parameter that has been widely used for detecting anomalies in a data set.
As per claim 11, Hsu-Kirchner teaches the method according to claim 1, Kirchner further teaches wherein the reconstruction error for individual dimensions is determined separately (see page 153 Col 1 “The reconstruction error e of the test measurements calculated by using the formula e j = Ix j − x0 jI (1) determines a value to describe the reconstruction at each specific point of a recording. By using a threshold tr, every measurement that exceeds the threshold is marked as an anomaly (e j > tr)…”, individual dimension has been interpreted as individual measurements points) and/or based on a metric and/or by means of a fast Fourier transformation and/or a wavelet transformation.
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu’s invention to include wherein the reconstruction error for individual dimensions is determined separately and/or based on a metric and/or by means of a fast Fourier transformation and/or a wavelet transformation as taught by Kirchner in order to determine anomalies in the laser process of Hsu using an autoencoder and detect anomalies exactly at the exact position of the anomalous pattern by performing the reconstruction error for individual dimensions separately (see page 153 Col 1 “…The main advantage of the reconstruction model is that it is easily possible to determine the exact position of the anomalous pattern within a time series. If a complete measurement consisting of up to 100.000 data points, is checked for anomalies, the measurement has to be split into subsequences because the input length of the LSTM cell is limited).
As per claim 15, Hsu-Kirchner teaches a laser machining method, comprising the steps of: Hsu further teaches machining a workpiece by means of a laser beam (see Fig. 1 machine 200 machining workpiece 206; also, see [0066] “ an example welding system 200 is shown in which a robot 202 welds a work piece 206 using a welding tool 208 (or, when under manual control, a handheld torch) to which power or fuel is delivered by welding equipment 210 via conduit 218 (for electrical welding, ground conduit 220 provides the return path…”; [0069] “The welding system 200 of FIG. 2 may be configured to cut material (e.g., as a plasma cutter) or form a solder or brazed joint, form a weld (e.g., weld joint 212) between two components in a weldment by any known welding technique, including Laser welding...); and
monitoring the laser machining process according to the method according to claim 1 (see Hsu and Kirchner teach the method of claim 1, same rationale applies herein).
As per claim 16, Hsu-Kirchner teaches a system for monitoring a laser machining process (see Hsu Fig. 1 and 4a-4e), said system comprising:
at least one sensor assembly configured to sense process signals of the laser machining process (see Hsu [0071] “As illustrated, the one or more sensors 236 may be positioned throughout the welding station (a/k/a weld cell) to measure and collect welding data, which may be used for weld production knowledge purposes. For instance, the one or more sensors may be positioned adjacent the weld and/or operable to capture one or more attributes of a weld (e.g., physical features of the weld) and/or one or more parameters of the weld (e.g., settings used when forming the weld), whether during fabrication of the weld or upon completion of the weld. The one or more sensors or transducers 236 may include any sensor useful in identifying defects, or measuring attributes/parameters, of a weld in a weldment. Examples of suitable sensors include, without limitation, current/LEM sensor, voltage and power sensors/calorimeter, encoders, photodiodes, cameras, microphones, seam finders, temperature sensors (e.g., positioned inside the welding equipment 210, or on the work piece 206), infrared (IR) detectors, proximity sensors, laser ranging and scanning devices, pressure sensors, inertial sensors, humidity sensors, airflow sensors, inertial measurement unit (IMU) sensors, shape memory alloy (SMA) sensors, piezoelectric sensors, nanotechnology chemical sensors, EMAT sensors, MEMS sensors, GPS, etc. In certain aspects, some of the one or more sensors 236 may be integral with, or coupled to, the welding equipment 210 and configured to measure and/or provide, for example, the arc current, wire drive current, arc voltage, power source input line voltage, tracker output, switch mode power source, pulse width modulation pulse width, or another parameter of the welding equipment 210. In certain aspects, as will be discussed below, the output from the one or more sensors 236 may be stored to a non-transitory medium for later analysis, where the data may be communicated in batches to a remote analytics system...”);
at least one autoencoder formed by a deep neural network (see [0087] “…unsupervised learning such as self-organizing maps and competitive layers; and deep learning such as convolutional neural net and auto encoders)…”);
and at least one processor configured to carry out the method for monitoring the laser machining process according to claim 1 (see [0110] “An exemplary analytics computing platform 234 may comprise a processor configured to perform one or more algorithms (e.g., weld production knowledge machine learning algorithms) and a non-transitory data storage device. The processor may be communicatively and operatively coupled with one or more non-transitory data storage devices, which may be a non-transitory, computer-readable medium having one or more databases (e.g., weld data store(s) having a large scale dataset) and/or computer-executable instructions embodied therein…”; also, see Hsu and Kirchner teach the method of claim 1, same rationale applies herein).
As per claim 17, Hsu-Kirchner teaches a laser machining system for machining a workpiece by means of a machining laser beam, said laser machining system comprising (see claim 1 above and see Hsu [0069] “laser welding”):
a laser machining head for radiating the machining laser beam onto said workpiece (see Hsu Fig. 2 work piece 206 and see [0069] “The welding system 200 of FIG. 2 may be configured to cut material (e.g., as a plasma cutter) or form a solder or brazed joint, form a weld (e.g., weld joint 212) between two components in a weldment by any known welding technique, including…laser welding…”); and
a system according to claim 16 (see Hsu and Kirchner teach the system of claim 16 and the method of claim 1, same rationale applies herein).
Claim(s) 4 and 6 are rejected under 35 U.S.C. 103 as being unpatentable over Hsu (US 20170032281) in view of Kirchner et al (“Detecting process instabilities in industrial gas metal arc welding time series”, 2020) as applied to claim 3 above, and further in view of Chen et al (“Real-time monitoring of high-power disk laser welding based on support vector machine, 2018”).
As per claim 4, Hsu-Kirchner teaches the method according to claim 3, while Hsu teaches or suggests the steps of evaluating the determined quality feature and suggests that machine learning algorithms are used for said purpose, Hsu dos not explicitly teach wherein: the step of determining a quality feature of the laser machining process is by means of a regressor formed by a neural network, and a value for the quality feature is determined, and/or the step of determining a quality feature of the laser machining process is by means of a classifier formed by a neural network, and a classification value for the quality feature is determined.
However, Chen teaches a method and system comprising determining a quality feature of the laser machining process, wherein: the step of determining a quality feature of the laser machining process is by means of a regressor formed by a neural network, and a value for the quality feature is determined (see the Abstract “…Fifteen features of laser-induced metal vapor plume and spatters were extracted and support vector machine was adopted to establish a classifier to evaluate the welding quality. Feature selection method was employed to choose suitable features…”, an SVM can be a regressor; also, see page 75 Col 2 “..As an increasingly popular tool for classification and regression [9], SVM has been widely applied in modern welding industry…”), and/or the step of determining a quality feature of the laser machining process is by means of a classifier formed by a neural network, and a classification value for the quality feature is determined (see the Abstract “…Fifteen features of laser-induced metal vapor plume and spatters were extracted and support vector machine was adopted to establish a classifier to evaluate the welding quality. Feature selection method was employed to choose suitable features…”, and SVM is a classifier; also, see page 75 Col 2 “…You et al. [11] introduced an innovative approach to perform laser welding process monitoring and weld defect diagnosis, SVM classification model was built to effectively identify weld defects. Mekhalfa and Nacereddine [12] used SVM to automatically classify four types of weld defects in radiographic images. Nevertheless, the performance of SVM is highly effected by the features used. Lack of precise priori knowledge and choosing features blindly may result in redundant or irrelevant features”; also, see page 80 Col 1-2 “…In machine learning and pattern recognition, SVM is widely used for classification. Experiment result shows that the SVM classifier generated by seven selected features can reach a satisfactory accuracy of 95.93% by 10-fold cross validation. This method is proven to be practical and can be applied in the real- time monitoring of high-power laser welding….”).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu-Kirchner’s combination to include a regressor or classifier, wherein: the step of determining a quality feature of the laser machining process is by means of a regressor formed by a neural network, and a value for the quality feature is determined, and/or the step of determining a quality feature of the laser machining process is by means of a classifier formed by a neural network, and a classification value for the quality feature is determined as taught by Chen in order to classify different type or class of welding defects with high accuracy (see page 75 Col 2 “…SVM to automatically classify four types of weld defects in radiographic images…”; also, see page 79 Col 2 “..Wrapper model was applied while searching for the optimal feature subset and the classification accuracy of the SVM classifier by 10-fold cross validation was used as the evaluation criterion. Complete search policy was employed to guarantee that the most suitable feature subset could be found, which meant the search would stop while all the candidate feature subsets had been tested… Performances of seven feature sets, including the six highest classification accuracy feature subsets and the entire data set (i.e. without feature selection), are shown in Table 3. From the table, it can be seen that the accuracy of class R is satisfactory with or without feature selection, which means the welding process of high welding quality is easier to identify and monitor due to its stability. As for class E, the accuracy cannot reach 50% while fifteen features are used, which indicates a much more complicated situation and the difficulty in recognizing the process of inferior welding quality. When feature selection is applied, six, seven or eight features are removed, the performance is improved significantly. The highest classification accuracy is 92.20%, which is more than twice of using the entire feature set, and only less than half of the features is used, which means the computational cost must be much lower”).
As per claim 6, Hsu-Kirchner-Chen the method according to claim 4, Hsu further teaches determining the quality feature based on the at least one process signal data set (see claim 3 above and see [0060], [0073], [0074], [0079], [0087], and [0093])
Hsu does not explicitly teach:
wherein: said autoencoder and at least one of said regressor and said classifier are parallel to each other (see Fig. 5 parallel );
and/or said regressor and/or said classifier determines the quality feature based on the at least one process signal data set (see );
and/or said autoencoder and at least one of said regressor and said classifier have a common input layer (see common input layer).
Chen further teaches a method and system comprising determining a quality feature of the laser machining process using a regressor and/or classifier, said regressor and/or said classifier determines the quality feature based on at least one process signal data set (see the Abstract “…Fifteen features of laser-induced metal vapor plume and spatters were extracted and support vector machine was adopted to establish a classifier to evaluate the welding quality. Feature selection method was employed to choose suitable features…”, an SVM can be a regressor; also, see page 75 Col 2 “..As an increasingly popular tool for classification and regression [9], SVM has been widely applied in modern welding industry…”; also, see page 79 Col 1 “…A fifteen-dimensional feature vector was generated by each image and thus a 1500 _ 15 data set is obtained. These 1500 samples from the data set were classified according to the welding quality…” page 79 Col 2 “The basic idea of 10-fold cross validation could be simply described as: the entire data set was equally divided into 10 parts, 9 of them were used to establish the SVM classifier and the rest one was for testing.).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu-Kirchner-Chen’s combination to include determining a quality feature of the laser machining process using a regressor and/or classifier, said regressor and/or said classifier determines the quality feature based on at least one process signal data set as taught by Chen in order to classify different type or class of welding defects with high accuracy (see page 75 Col 2 “…SVM to automatically classify four types of weld defects in radiographic images…”; also, see page 79 Col 2 “..Wrapper model was applied while searching for the optimal feature subset and the classification accuracy of the SVM classifier by 10-fold cross validation was used as the evaluation criterion. Complete search policy was employed to guarantee that the most suitable feature subset could be found, which meant the search would stop while all the candidate feature subsets had been tested… Performances of seven feature sets, including the six highest classification accuracy feature subsets and the entire data set (i.e. without feature selection), are shown in Table 3. From the table, it can be seen that the accuracy of class R is satisfactory with or without feature selection, which means the welding process of high welding quality is easier to identify and monitor due to its stability. As for class E, the accuracy cannot reach 50% while fifteen features are used, which indicates a much more complicated situation and the difficulty in recognizing the process of inferior welding quality. When feature selection is applied, six, seven or eight features are removed, the performance is improved significantly. The highest classification accuracy is 92.20%, which is more than twice of using the entire feature set, and only less than half of the features is used, which means the computational cost must be much lower”) and it would have been obvious to use the same process signal data of the System of Hsu in order to identify the same defects as in the system of HSU.
Claim(s) 7 is rejected under 35 U.S.C. 103 as being unpatentable over Hsu (US 20170032281) in view of Kirchner et al (“Detecting process instabilities in industrial gas metal arc welding time series”, 2020) and Chen et al (“Real-time monitoring of high-power disk laser welding based on support vector machine”, 2018), as applied to claim 3 above, and further in view of Rosenfeld et al (US 20210089842).
As per claim 7, Hsu-Kirchner-Chen teaches the method according to claim 4, but it does not explicitly teach wherein said autoencoder and at least one of said regressor and said classifier are trained with the same data.
However, Rosenfeld teaches a method for classifying sensor data comprising an autoencoder and at least one of said regressor and said classifier are trained with the same data (see [0037] “One may also opt to train the encoder together with a classifier on the same training data as used during inference…”; also, see Fig. 2 the encoder and classifier are trained with the same data in storage 210, see [0071], [0076] and see [0077-0080] “… In an embodiment, classifier 230 may be trained on the training data 210…”; also, see [0084], [0088]).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified an autoencoder and at least one of said regressor and said classifier are trained with the same data as taught by Rosenfeld in order to classify sensor/input data with improved robustness (see [0006]) and because training the encoder and regressor/classifier with the same data optimizes the feature extraction and classification process (this is common sense because the idea is to use two anomalies detection systems, wherein the encoder uses the data to identify an anomaly in a broad manner and the classifier determines a specific anomaly or class anomaly).
Claim(s) 9-10 and 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over Hsu (US 20170032281) in view of Kirchner et al (“Detecting process instabilities in industrial gas metal arc welding time series”, 2020) as applied to claim 3 above, and further in view of Malhotra et al (US 20200012918).
As per claim 9, Hsu-Kirchner teaches the method according to claim 1, but it does not explicitly teach
wherein the step of determining a reconstruction error comprises
determining a Mahalanobis distance with respect to:
a deviation of the at least one process signal data set from the at least one reconstructed process signal data set;
and/or individual characteristic values of the reconstruction error;
and/or encoding of a process signal data set.
Malhotra teaches an anomaly detection method and system comprising wherein the step of determining a reconstruction error comprises determining a Mahalanobis distance with respect to (see Fig. 3a): a deviation of the at least one process signal data set from the at least one reconstructed process signal data set (see Fig. 3A and see [0033] Given the error vector e.sub.t.sup.(i), Mahalanobis distance is used to compute the anomaly score a.sub.t.sup.(i) as follows: a.sub.t.sup.(i)=√{square root over ((e.sub.t.sup.(i)−μ).sup.Tϵ.sup.−1(e.sub.t.sup.(i)−μ))}(2)), and/or individual characteristic values of the reconstruction error (see Fig. 3A), and/or encoding of a process signal data set (see Fig. 3A).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu-Kirchner’s combination as taught above to include wherein the step of determining a reconstruction error comprises determining a Mahalanobis distance with respect to: a deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or individual characteristic values of the reconstruction error; and/or encoding of a process signal data set as taught by Malhotra in order to generate an anomaly score and detect of classify an anomaly in a system process (see [0034], 0041 “… the step of classifying at least one time instance in the multi-dimensional time series as anomalous if the anomaly score is greater than a threshold. In an embodiment, this approach further includes the step of classifying at least one time instance in the multi-dimensional time series as normal if the anomaly score is less than or equal to the threshold. In an embodiment, F-score corresponding to a binary classifier with two classes i.e. a normal class (0) and an anomalous class (1)”).
As per claim 10, Hsu-Kirchner teaches the method of claim 8, but it does not explicitly teach wherein said method comprises determining parameters mean vector and covariance matrix of the Mahalanobis distance using defect-free or labeled data sets.
However, Malhotra teaches an anomaly detection method and system comprising wherein the step of determining a reconstruction error comprises determining a Mahalanobis distance (see Fig. 3A and see [0033] Given the error vector e.sub.t.sup.(i), Mahalanobis distance is used to compute the anomaly score a.sub.t.sup.(i) as follows: a.sub.t.sup.(i)=√{square root over ((e.sub.t.sup.(i)−μ).sup.Tϵ.sup.−1(e.sub.t.sup.(i)−μ))}(2)), determining parameters mean vector and covariance matrix of the Mahalanobis distance using defect-free or labeled data sets (see [0034] where μ and Σ are the mean and covariance matrix of the error vectors corresponding to the normal training time series instances. This anomaly score can be obtained in an online setting by using a window of length T ending at current time t as the input, making it possible to generate timely alarms related to anomalous behavior….”, defect free data sets has been interpreted as normal data set; also, see [0052]).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu-Kirchner’s combination as taught above to include determining parameters mean vector and covariance matrix of the Mahalanobis distance using defect-free or labeled data sets as taught by Malhotra in order to generate an anomaly score and detect of classify an anomaly in a system process (see [0034], 0041 “… the step of classifying at least one time instance in the multi-dimensional time series as anomalous if the anomaly score is greater than a threshold. In an embodiment, this approach further includes the step of classifying at least one time instance in the multi-dimensional time series as normal if the anomaly score is less than or equal to the threshold. In an embodiment, F-score corresponding to a binary classifier with two classes i.e. a normal class (0) and an anomalous class (1)”).
As per claim 13, Hsu-Kirchner teaches the method according to claim 1, but it does not exactly teach wherein said method comprises: determining a degree of abnormality; wherein detecting an anomaly of the laser machining process is based on the determined degree of abnormality.
Malhotra teaches an anomaly detection method and system comprising determining a degree of abnormality (see Fig. 3a ), wherein detecting an anomaly of the laser machining process is based on the determined degree of abnormality (see Fig. 3A and see [0033] Given the error vector e.sub.t.sup.(i), Mahalanobis distance is used to compute the anomaly score a.sub.t.sup.(i) as follows: a.sub.t.sup.(i)=√{square root over ((e.sub.t.sup.(i)−μ).sup.Tϵ.sup.−1(e.sub.t.sup.(i)−μ))}(2); also, see [0034 and [0041]).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu-Kirchner’s combination as taught above to include determining a degree of abnormality; wherein detecting an anomaly of the laser machining process is based on the determined degree of abnormality as taught by Malhotra in order to generate an anomaly score and detect of classify an anomaly in a system process (see [0034], 0041 “… the step of classifying at least one time instance in the multi-dimensional time series as anomalous if the anomaly score is greater than a threshold. In an embodiment, this approach further includes the step of classifying at least one time instance in the multi-dimensional time series as normal if the anomaly score is less than or equal to the threshold. In an embodiment, F-score corresponding to a binary classifier with two classes i.e. a normal class (0) and an anomalous class (1)”).
As per claim 14, Hsu-Kirchner-Malhotra teaches the method according to claim 13, Malhotra further teaches wherein the determining a degree of abnormality is based on a weighted summation or on a Mahalanobis distance with respect to individual characteristic values for the reconstruction error (see Fig. 3A and see [0033] Given the error vector e.sub.t.sup.(i), Mahalanobis distance is used to compute the anomaly score a.sub.t.sup.(i) as follows: a.sub.t.sup.(i)=√{square root over ((e.sub.t.sup.(i)−μ).sup.Tϵ.sup.−1(e.sub.t.sup.(i)−μ))}(2); also, see [0034] where μ and Σ are the mean and covariance matrix of the error vectors corresponding to the normal training time series instances. This anomaly score can be obtained in an online setting by using a window of length T ending at current time t as the input, making it possible to generate timely alarms related to anomalous behavior. A point x.sub.t.sup.(i) is classified as anomalous if a.sub.t.sup.(i)>τ; the threshold τ can be learned using a hold-out validation set while optimizing for F-score and [0041] ).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu-Kirchner’s combination as taught above to include wherein the determining a degree of abnormality is based on a weighted summation or on a Mahalanobis distance with respect to individual characteristic values for the reconstruction error as taught by Malhotra in order to generate an anomaly score and detect of classify an anomaly in a system process (see [0034], [0041] “… the step of classifying at least one time instance in the multi-dimensional time series as anomalous if the anomaly score is greater than a threshold. In an embodiment, this approach further includes the step of classifying at least one time instance in the multi-dimensional time series as normal if the anomaly score is less than or equal to the threshold. In an embodiment, F-score corresponding to a binary classifier with two classes i.e. a normal class (0) and an anomalous class (1)”).
Claim(s) 12 is rejected under 35 U.S.C. 103 as being unpatentable over Hsu (US 20170032281) in view of Kirchner et al (“Detecting process instabilities in industrial gas metal arc welding time series”, 2020) as applied to claim 1 above, and further in view of Naresh et al ( EP 3878583 A1).
As per claim 12, Hsu-Kirchner teaches the method according to claim 1, while Hsu teaches the data is filtered (see 0083 and 0100), Hsu-Kirchner does not explicitly teach
wherein: said method comprises normalizing the reconstruction error with respect to the process signal data set;
and/or the step of determining a reconstruction error comprises:
filtering at least part of the process signal data set and/or the reconstructed process signal data set;
and based thereon, determining the reconstruction error.
Naresh teaches a method and system for monitoring a process comprising wherein: said method comprises normalizing the reconstruction error with respect to the process signal data set; and/or the step of determining a reconstruction error comprises (see Fig. 3 and Fig. 4A reconstruction): filtering at least part of the process signal data set and/or the reconstructed process signal data set (see Fig. 3 filter 320); and based thereon, determining the reconstruction error (see Fig. 3 and see [0065] “…An
image filter 320 processes the image data from the database 310 to filter dark frames of the image data, identify label(s) 325 in the image frames/slices, etc. For example, images of example, actual, and/or reference melt pools
are filtered to remove dark frames of melt pool image 20 can be deployed.
data (e.g., frames that do not include an image of the melt pool and are instead all or mostly "dark") and identify label(s) 325 in the remaining melt pool image data…”; also, see [0095] and [0101] and [0117] “…to filter the image data before extracting the image features… Example 10 includes the method of any preceding
clause, further including filtering the image data before applying the artificial intelligence model”).
Therefore, it would have been obvious to one of ordinary skilled in the art before effective filing date of the claimed invention to which said subject matter pertains to have modified Hsu-Kirchner’s combination as taught above to include wherein: said method comprises normalizing the reconstruction error with respect to the process signal data set; and/or the step of determining a reconstruction error comprises: filtering at least part of the process signal data set and/or the reconstructed process signal data set, and based thereon, determining the reconstruction error as taught by Naresh in order to clean the data set or remove noise from the data set before is input into the autoencoder model to avoid false negatives.
Conclusion
The prior art made of record and not relied upon, as cited in PTO form 892, is considered pertinent to applicant's disclosure.
Zhang et al, "Data-Driven Detection of Laser Welding Defects Based on Real-Time Spectrometer Signals, 2019” teaches an autoencoder to detect anomalies and a classifier to detect anomalies (fig. 2) in laser welding system process (page 9371 col 2).
Wikipedia “autoencoder” teaches how encoders are structured and the purpose which is used to detect anomalies in a dataset based on a reconstruction error (see page 9 anomaly detection).
Zhang et al, "A Low-Cost Welding Status Monitoring Framework for High-Power Disk Laser Welding (December 2018),” teaches a system and method for detecting anomalies for a laser welding process comprising an autoencoder (see Fig. 1 and 2) and a classifier (fig. 3) and for classifying the anomalies.
Zhang et al, "Real-time monitoring of high-power disk laser welding statuses based on deep learning framework", 2020”, teaches an autoencoder to detect anomalies combined with a classifier to classify anomalies (fig. 9) in laser welding system process (page 807-808).
Syngal et al (US 20220374684) teaches an autoencoder combined with a classifier (see fig. 5).
Allahdadian et al (US 20220156578) teaches a reconstruction model such as an autoencoder, determining a reconstruction errors to detect anomalies in data, and using normalized errors, calculating a mahala Nobis distance, to detect anomalies (0093-0085).
Gregson et al (US 20210342570) teaches an autoencoder combined with a classifier (see fig. 5 and 0035-0036).
Jun et al (US 20210256991) teaches an autoencoder combined with a classifier to detect anomalies (see fig. 8 and 0057, 0060, and 0065).
Fan et al (US 20210049452) teaches an autoencoder combined with a classifier to detect anomalies (see fig. 2A and 0023, 0032).
Dong et al (US 20200210808) teaches an autoencoder combined with a classifier to detect anomalies (see fig. 2).
Examiner respectfully requests, in response to this Office action, support be shown for language added to any original claims on amendment and any new claims. That is, indicate support for newly added claim language by specifically pointing to page(s) and line number(s) in the specification and/or drawing figure(s). This will assist Examiner in prosecuting the application.
When responding to this Office Action, Applicant is advised to clearly point out the patentable novelty which he or she thinks the claims present, in view of the state of the art disclosed by the references cited or the objections made. Applicant must also show how the amendments avoid or differentiate from such references or objections. See 37 CFR 1.111 (c).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to OLVIN LOPEZ ALVAREZ whose telephone number is (571) 270-7686 and fax (571) 270-8686. The examiner can normally be reached Monday thru Friday from 9:00 A.M. to 6:00 P.M.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Robert Fennema, can be reached at (571) 272-2748. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center. Status information for published applications may be obtained from Patent Center. Status information for unpublished applications is available through Patent Center for authorized users only. Should you have questions about access to Patent Center, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
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) Form at https://www.uspto.gov/patents/uspto-automated- interview-request-air-form.
/O. L./
Examiner, Art Unit 2117
/ROBERT E FENNEMA/Supervisory Patent Examiner, Art Unit 2117
Prosecution Insights