DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Notice to Applicant
The following is a Non-Final Office action, responsive to Applicant’s communication of 6/7/23, in which Applicant filed the application. Claims 1-20 are pending in this application and have been rejected below.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 6/7/23, 12/20/24, 9/11/25, and 4/24/26 are being considered by the examiner.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are:
a first data acquisition part in claim 1, 12, 15;
a second data acquisition part in claim 1, 12, 15;
a data transfer part in claim 1, 12;
a machine learning part in claim 1, 12, 15;
a specifying part in claim 1, 15;
an output part in claim 1, 15;
a confirmation part in claim 2;
a candidate presenting part in claim 8.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
The corresponding structure, based on [0055-0057] as published and FIG. 3 is a CPU executing stored instructions.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
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 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e. an abstract idea) without reciting significantly more.
Step One - First, pursuant to step 1 in MPEP 2106.03, the claim 1 is directed to a system which is a statutory category.
Step 2A, Prong One - MPEP 2106.04 - The claim 1 recites–
A quality prediction …, comprising:
a first data acquisition … configured to acquire first learning data which is collected in first manufacturing equipment and indicates a relationship between each adjustment item of the first manufacturing equipment and a quality index of a first product manufactured by the first manufacturing equipment (Applicant’s [0049] as published gives various examples of adjustment items during manufacture of products such as “processing speed, processing temperature, room temperature, position of device, angle of robot arm, torque, pressure, flow rate, input angle/position/amount of material, current/voltage, frequency, and various control parameters (PID control, etc.).”);
a second data acquisition … configured to acquire second learning data which is collected in second manufacturing equipment and indicates a relationship between each adjustment item of the second manufacturing equipment and a quality index of a second product manufactured by the second manufacturing equipment;
a data transfer … configured to convert the acquired first learning data to match the second learning data to implement transfer … (Applicant’s specific examples of “convert” from first manufacturing equipment data to a second equipment is [0078] converting using a “normalization/standardization”; and [0092] as published where its alignment using a mathematical covariance matrix. Either way, the “convert” appears to be a normalization to get the 1st data to fit the scale or environment of the 2nd equipment);
… configured to implement … a quality prediction model for predicting the quality index of the second product from the each adjustment item of the second manufacturing equipment, using the converted first learning data and the second learning data;
a specifying … configured to specify an adjustment amount of the each adjustment item of the second manufacturing equipment so that the predicted quality index of the second product satisfies a quality standard, using the … quality prediction model; and
an output … configured to output the specified adjustment amount of the each adjustment item of the second manufacturing equipment.
As drafted, this is, under its broadest reasonable interpretation, within the Abstract idea grouping of “mathematical relationships” AND “certain methods of organizing human activity” (managing personal behavior or relationships or interactions between people (including following rules or instructions), as the claim involves acquiring 1st manufacturing data with relationship of “adjustment data” to product quality, acquiring 2nd manufacturing data with relationships of “adjustment data” to second product quality; converting data from 1st equipment to 2nd equipment (which Applicant’s examples are normalization or alignment in [0078, 0092), predicting quality index for using converted adjustment item on the 2nd equipment to satisfy a quality standard, and outputting, presumably to a person, the adjustment amount as a recommendation instruction to a person. The “acquiring” of the data from 1st or 2nd equipment may be done “manually” by an operator (See Applicant’s [0070, 0074] as published). Accordingly, at this time, claim 1 is directed to an abstract idea, as it is for analyzing some data for a 1st manufacturing equipment related to quality, predicting quality for 2nd manufacturing equipment based on data converted from the 1st equipment, and outputting what adjustment data to satisfy a quality standard on second manufacturing equipment. The claim is also directed to mathematical relationships as it “acquires” a relationship between “adjustment item” and a quality index for a first equipment, acquires data for a second manufacturing equipment, converts [where examples in 0078, 0092 are math] from 1st to 2nd equipment, predicts quality index for 2nd product from each adjustment, specify an adjustment amount based on predicted quality satisfying a quality standard (e.g. a number threshold).
Step 2A, Prong Two - MPEP 2106.04 - This judicial exception is not integrated into a practical application. In particular, the claim recites additional elements that are:
A quality prediction system, comprising:
a first data acquisition part configured to acquire first learning data which is collected in first manufacturing equipment and indicates a relationship between each adjustment item of the first manufacturing equipment and a quality index of a first product manufactured by the first manufacturing equipment;
a second data acquisition part configured to acquire second learning data which is collected in second manufacturing equipment and indicates a relationship between each adjustment item of the second manufacturing equipment and a quality index of a second product manufactured by the second manufacturing equipment;
a data transfer part configured to convert the acquired first learning data to match the second learning data to implement transfer learning (Applicant’s specification [0088] states “A known transfer learning method may be employed as the conversion method. In one example, the controller 11 may convert the first learning data 30 to match the second learning data 35 by the Frustratingly Easy Domain Adaptation method or the Correlation Alignment method.”);
a machine learning part configured to implement machine learning of a quality prediction model for predicting the quality index of the second product from the each adjustment item of the second manufacturing equipment, using the converted first learning data and the second learning data;
a specifying part configured to specify an adjustment amount of the each adjustment item of the second manufacturing equipment so that the predicted quality index of the second product satisfies a quality standard, using the trained quality prediction model; and
an output part configured to output the specified adjustment amount of the each adjustment item of the second manufacturing equipment.
The claims, individually or when viewed in combination, are viewed reciting the computer at a high-level of generality (i.e., as a generic processor performing each step) such that it amounts no more than mere instructions to apply the exception using a generic computer component. See MPEP 2106.05(f) and “field of use” (MPEP 2106.05h) for the “machine learning,” “transfer learning,” and computer required by the “part” [as interpreted under 112f]. The claim here just “uses machine learning” to output recommended adjustments to a person. At this time, the claims do not, individually or in combination, reflect a technical improvement. Accordingly, the additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim also fails to recite any improvements to another technology or technical field, improvements to the functioning of the computer itself, use of a particular machine, effecting a transformation or reduction of a particular article to a different state or thing, and/or an additional element applies or uses the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception. See 84 Fed. Reg. 55. The claim is directed to an abstract idea.
Step 2B in MPEP 2106.05 - The claim does 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 in claim 1 of “computer” [from 112 interpretation of “part”] and “machine learning” limitations are “apply it” [abstract idea] on a computer. (See MPEP 2106.05(f) – Mere Instructions to Apply an Exception – “Thus, for example, claims that amount to nothing more than an instruction to apply the abstract idea using a generic computer do not render an abstract idea eligible.” Alice Corp., 134 S. Ct. at 235) and “field of use” (MPEP 2106.05h). Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim fails to recite any improvements to another technology or technical field, improvements to the functioning of the computer itself, use of a particular machine, effecting a transformation or reduction of a particular article to a different state or thing, adding unconventional steps that confine the claim to a particular useful application, and/or meaningful limitations beyond generally linking the use of an abstract idea to a particular environment. See 84 Fed. Reg. 55. The claim is not patent eligible. Viewed individually or as a whole, these additional claim element(s) do not provide meaningful limitation(s) to transform the abstract idea into a patent eligible application of the abstract idea such that the claim(s) amounts to significantly more than the abstract idea itself.
Independent claim 12 is directed to a device at step 1, which is a statutory category. Claim 12 recites similar limitations as claim 1, though claim 12 is broader as it does not include the last two steps of claim 1. Accordingly, claim 12 is rejected for the same reasons at step 2A, prong one, step 2A, prong 2, and step 2B. The claim is not patent eligible.
Independent claim 13 is directed to a method at step 1, which is a statutory category. Claim 13 recites similar limitations as claim 1. Claim 13 recites “computer to execute” in preamble, as opposed to “part[s]” interpreted under 112f as requiring a computer. The result under 101 is the same. Accordingly, claim 13 is rejected for the same reasons at step 2A, prong one, step 2A, prong 2, and step 2B. The claim is not patent eligible.
Independent claim 14 is directed to an article of manufacture at step 1, which is a statutory category. Claim 14 recites similar limitations as claim 1, though claim 14 recites “non-transient computer-readable recording medium recording a program for a computer to execute” each limitation. These are additional elements analyzed at step 2A, prong 2 and step 2B; individually or in combination, they are viewed as “apply it [abstract idea] on a computer” (MPEP 2106.05f), and “field of use” (MPEP 2106.05h) for specific medium storing program executed by a computer, as above. Accordingly, claim 14 is rejected for the same reasons at step 2A, prong one, step 2A, prong 2, and step 2B. The claim is not patent eligible.
Independent claim 15 is directed to a system at step 1, which is a statutory category. Claim 15 recites similar limitations as claim 1 and is rejected for the same reasons at step 2A, prong one, step 2A, prong 2, and step 2B. At step 2A, prong one, claim 15 also recites “kernel function” and “weighting” which is viewed as narrowing the abstract idea with a further mathematical relationship. At step 2a, prong two and step 2B, having “machine learning constructing a kernel function” is viewed as “apply it” on a computer (MPEP 2106.05f) and “field of use” (MPEP 2106.05h). The claim is not patent eligible.
Claim 2 narrows the abstract idea by mathematically evaluating whether an adjustment item statistically does not contribute to predicting quality from either the first or second data acquired..
Claim 3 narrows the abstract idea by acquiring additional second data by experimenting under a plurality of conditions. Similar to claim 1 above, the claim 3 also has additional element of computer, and machine learning which are viewed as “apply it [abstract idea] on a computer” MPEP 2106.05(f) and “field of use” (MPEP 2106.05h).
Claim 4-5, 16-17 narrow the abstract idea by having a series of mathematical relationships as part of the statistically analyzing limitation of claim 2.
Claims 6, 18-20 narrow the abstract idea by stating the named algorithm is either “Frustratingly Easy Domain Adaptation” or “Correlation Alignment.”
Claim 7-8 narrows the abstract idea by acquiring additional second data, then predict quality, then predict an adjustment amount. Similar to claim 1 above, the claim 7, 8 also has additional element of computer, and machine learning, and “retrained” (as in do the same steps, based on additional information), and claim 8 “presents” information (i.e. displays, presumably to a person), which are viewed as “apply it [abstract idea] on a computer” MPEP 2106.05(f) and “field of use” (MPEP 2106.05h).
Claim 9 narrows the abstract idea by stating that the data acquired represents data from a “basic” production line, and the second equipment is a “duplicate” production line.
Claim 10 narrows the abstract idea by stating that the data acquired represents data from a “production line” before making a change, and the second equipment is a after marking a change.
Claim 11 has additional elements by stating that the quality prediction model is configured by either a neural network, a support vector machine. The alternative of a “regression model” is viewed as narrowing the abstract idea, by having a mathematical model.
Therefore, the claim(s) are rejected under 35 U.S.C. 101 as being directed to non-statutory subject matter.
For more information on 101 rejections, see MPEP 2106.
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 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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-3 and 7-14 are rejected under 35 U.S.C. 103 as being unpatentable over Putman (US 2020/0401119) in view of Liu et al., "Domain adaptation transfer learning soft sensor for product quality prediction," 2019, Chemometrics and Intelligent Laboratory Systems, Vol. 192, No. 103813, pages 1-11.
Concerning claim 1, Putman discloses:
A quality prediction system (Putman – see par 5 - a system having one or more processors; see par 21 - Each process station can include one or more tools/equipment that performs a set of process steps on: received raw materials (this can apply to a first station or any of the subsequent stations in the manufacturing process) and/or the received output from a prior station (this applies to any of the subsequent stations in the manufacturing process). Examples of process stations can include… quality control; see par 37 - FIG. 2 illustrates an example deep learning controller 218 that can be configured to control any number of (referred to herein by “n”) processing stations in a manufacturing process using predictive process control, as discussed later in connection with FIGS. 4-7; see par 56 - deep learning controller 218 can predict the values for the characteristics of the final output (“expected value” or “EV”) that determine whether or not the final output value will be acceptable or not (i.e., whether or not the final output is “in specification”) (step 342)), comprising:
a first data acquisition part (corresponding structure under 112f, based on [0055-0057] as published and FIG. 3 is a CPU executing stored instructions – Putman discloses – see par 5 - a system having one or more processors, and a non-transitory memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations; see par 96 - Deep learning controller 218 can include a processing unit (e.g., CPU/s and/or processor/s) 910 ) configured to acquire first learning data which is collected in first manufacturing equipment (Putman- see par 51 - As the process iterates through each station, all the values associated with: an individual station (e.g., control values); an output of an individual station (e.g., station values, intermediate/final output values), or multiple stations (e.g., process values) are measured or calculated and provided to condition the machine learning algorithms of deep learning controller 318 (steps 327 and 328). see par 52 - manufacturing performance metrics (e.g., … percentage of products not in specification for a specified time period,…) for the manufacturing process under conventional control can be calculated and provided to deep learning controller 218 (step 329). see par 55 - deep learning controller 218 draws inferences simply by analyzing the received data that it collects during the iteration of the manufacturing process (e.g., steps 328 and 329).) and indicates a relationship between each adjustment item of the first manufacturing equipment and a quality index of a first product manufactured by the first manufacturing equipment (Applicant’s [0049] as published gives various examples of adjustment items during manufacture of products such as “processing speed, processing temperature, room temperature, position of device, angle of robot arm, torque, pressure, flow rate, input angle/position/amount of material, current/voltage, frequency, and various control parameters (PID control, etc.).”); Putman discloses the limitations based on broadest reasonable interpretation in light of the specification – see par 53 - Any actions taken (or control signals generated) by the station controller in response to a received control value or other control input from a process station can be provided to deep learning controller 218. Such actions can include adjusting temperature, speed, etc. In addition, deviations from: acceptable setpoints, acceptable intermediate/final output values, acceptable control/station/process values can also be calculated and provided to deep learning controller 218. see par 59 - deep learning controller 218 can also, after initialization of the station controllers, at the outset of an iteration of the manufacturing process (i.e., proceeding through all stations in the manufacturing process), as well as over the course of the manufacturing process, predict whether any control inputs will cause unsatisfactory station performance or impact process performance (i.e., cause unacceptable process performance));
a second data acquisition part configured to acquire second learning data which is collected in second manufacturing equipment and indicates a relationship between each adjustment item of the second manufacturing equipment and a quality index of a second product manufactured by the second manufacturing equipment (Putman – see par 57 - deep learning controller 218 can be configured to perform EV (see par 56 – values for final output “Expected value”) predictions regarding output characteristics on a station-by-station basis. That is, deep learning controller 218 can make EV predictions regarding outputs at a specific station, and subsequently compare those predictions to actual outputs observed at that station. Alternatively, EV predictions can be made for outputs resulting from combined processing performed by two or more stations, depending on the desired implementation.);
a data transfer part configured to convert the acquired first learning data to match the second learning data to implement transfer learning (Putman – see par 55 - the training of deep learning controller 218 can be augmented by: providing deep learning controller 218 with simulated data or data from a similar manufacturing process. In one embodiment, deep learning controller 218 can be conditioned by implementing deep learning controller 218 into a similar manufacturing process and fine-tuning the deep learning controller during implementation in the target manufacturing process. That is, training of deep learning controller 218 can be performed using a training process that is performed before deep learning controller 218 is deployed into a target manufacturing environment.)
Putman discloses training using data from “similar” manufacturing (See par 55). Putman mentions in [0091] that its system optimization can reduce possible actions/states by incorporating by reference “62/836,213 “Transfer Learning Approach to Multi-Component Manufacturing Control”.
Liu discloses:
a data transfer part configured to “convert the acquired first learning data to match” the second learning data to implement “transfer learning” (Liu – see Abstract, “domain adaptation extreme learning machine (DAELM)” to establish sensor model suitable for multi-grade processes with limited labeled data; see page 2, col. 1, 1st paragraph - Transfer learning [36,37], as a branch of machine learning, can effectively deal with the distribution discrepancy of data among different domains; see page 4, 1st column - To construct the DAELM model, the empirical error from the target domain is introduced as the extra regularization term to learn its output matrix W Two ridge parameters λS > 0 and λT > 0
are carefully selected so that the source domain data are not overfitted
while the predictions of the target domain data are the main focus of this model.
The output matrixW, as a weighted combination between the source domain and the target domain, is calculated by solving the objective function in equation 6. Liu – see page 2, col. 1, 2nd -3rd paragraphs - In summary, transfer learning can improve the accuracy of a prediction model constructed using insufficient labeled data, by transferring the useful information from the labeled data belonging to
other related operating conditions. As a simple transfer learning method,
the domain adaptation ELM (DAELM) [38] is utilized in this paper to
construct a soft sensor model for the multi-grade chemical processes)
Putman and Liu disclose:
a machine learning part configured to implement machine learning of a quality prediction model for predicting the quality index of the second product from the each adjustment item of the second manufacturing equipment, using the converted first learning data and the second learning data (Putman – see par 36 - Machine learning models, as discussed herein, can also be used to determine the process stations, control/station/process values and intermediate output values that are most influential on the final output value (“key influencers”), and to optimize the manufacturing process by targeting the key influencers. see par 39 - Experiential priors, as used herein, refers to information gained by prior experience with, for example performing the same or similar manufacturing process; operating the same or similar stations; producing the same or similar intermediate/final outputs. In some embodiments, experiential priors can include acceptable final output values or unacceptable final output values. Acceptable final output values refer to an upper limit, lower limit or range of final output values where the final output is considered “in specification.” Conversely, unacceptable final output values refer to upper/lower limits or range of final output values where the final output is “not in specification” (i.e., describe the parameters for final output values that do not meet design specifications). Experiential priors can also include acceptable and unacceptable manufacturing performance metrics. see par 61 - Deep learning controller 218 uses its conditioned machine learning algorithms (as discussed in connection with FIG. 3) to calculate control inputs for the station controllers associated with the process stations of a manufacturing process. Based on the calculated control inputs, deep learning controller 218 can predict the expected value (EV) of the final output for the manufacturing process, along with a confidence level for its prediction (step 405). ). The machine learning algorithms continue to be improved throughout the implementation of PPC (predictive process control) (step 335). Further, the functional and experiential priors can be dynamically updated throughout PPC.);
a specifying part configured to specify an adjustment amount of the each adjustment item of the second manufacturing equipment so that the predicted quality index of the second product satisfies a quality standard, using the trained quality prediction model (Putman – see par 63 - shown in FIG. 4, deep learning controller 218 can predict the expected value (EV) for the final output, determine whether the expected value for the final output is in-specification, determine the confidence level for its prediction, and then provide feedback on its prediction by comparing the expected final value to the actual final value (step 445). Further, if deep learning controller 218 determines that the final output is not in-specification, it can calculate adjustments to the control inputs, so that the predicted expected value for the final output is in-specification. see par 83 - Deep learning controller 218 can provide the optimal control inputs to the relevant station controllers at the outset and continuously throughout the manufacturing process (step 620). The optimal control inputs do not have to be provided to the station controllers in serial order, but can be provided to one or more station controllers in parallel or in any order that is suitable to produce final outputs that achieve the desired optimal design or process values. see par 88 - In particular, based on information received from prior stations in a process, deep learning controller 218 can make changes to the control inputs associated with later stations in the process to ensure the optimal design/process values are achieved. Deep learning controller 218, can also adjust prior stations in the process as it proceeds through the process and receives data from subsequent stations;
see also Liu – see page 10, col. 1, 2nd paragraph - the transfer learning technique was introduced into the field of soft sensor development for multi-grade chemical processes. Specifically, a DAELM soft sensing method was proposed, which is able to transfer process information between different operating grades and
enhance the online quality predictions of the grades with limited labeled training data. Additionally, an FLOO strategy was developed to select the model parameters. the DAELM model can better capture useful information from various operating conditions and then enhance its prediction performance. The prediction results of two multi-grade chemical processes illustrate the advantages of DAELM.); and
an output part configured to output the specified adjustment amount of the each adjustment item of the second manufacturing equipment (Putman – see par 45 - Using predictive process control, as described herein, deep learning controller 218 can determine whether the wire was cut to the desired length specification and provide improvements to the cutting process, for example, that are provided in the form of a set of instructions to the operator of the manual station; see par 61 - deep learning controller 218 can output the calculated control inputs to the station controllers associated with the process stations of the manufacturing process (step 420). see par 67 - deep learning controller 218 can, not only initialize the station controller inputs before the start of an iteration through the process stations of the manufacturing process, but also adjust station controller inputs during the process itself (“feedforward control”). see par 98 - An output device 935 can also be one or more of a number of output mechanisms (e.g., printer, monitor) known to those of skill in the art. Data output from deep controller 218 can be displayed visually, printed, or generated in file form and stored in storage device 930 or transmitted to other components for further processing.)
Both Putman and Liu are analogous art as they are directed to predicting values/quality for manufacturing output/products (See Putman Abstract; Liu Abstract, par 57). Putman discloses training using data from “similar” manufacturing (See par 55). Putman mentions in [0091] that its system optimization can reduce possible actions/states by incorporating by reference “62/836,213 “Transfer Learning Approach to Multi-Component Manufacturing Control”. Liu improves upon Putman by disclosing using “transfer learning” and using mathematics so that the data from the target domain with operating conditions (e.g. a 1st set of manufacturing data) is not overfitted (e.g. a 2nd set of manufacturing data). One of ordinary skill in the art would be motivated to further include explicitly using “transfer learning” along with a process for using data from a target domain to a source domain to efficiently improve upon the use of data from “similar” manufacturing, where “Transfer Learning” is mentioned in Putman (See par 55, 91).
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system and method of using machine learning for adjusting control inputs if expected values are not in-specification in Putman (Abstract, par 63) to further use transfer learning for different types/grades of manufacture from different domains for quality prediction as disclosed in Liu, since the claimed invention is merely a combination of old elements, and in combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable and there is a reasonable expectation of success.
Concerning independent claim 12, Putman and Liu disclose:
A model-generating device (Putman see par 5 - a system having one or more processors, and a non-transitory memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations; see par 55-56 - Based on the conditioning of the machine learning models, deep learning controller 218 can predict the values for the characteristics of the final output (“expected value” or “EV”) that determine whether or not the final output value will be acceptable or not (i.e., whether or not the final output is “in specification”) (step 342); see par 59 – predict process performance; see par 96 - Deep learning controller 218 can include a processing unit (e.g., CPU/s and/or processor/s) 910 ), comprising:
The remaining limitations are similar to claim 1. Claim 12 is rejected for the same reasons as claim 1.
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1.
Concerning independent claim 13, Putman and Liu disclose:
A quality prediction method for a computer to execute (Putman see par 5 - a system having one or more processors, and a non-transitory memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations; see par 59 – predict process performance; see par 96 - Deep learning controller 218 can include a processing unit (e.g., CPU/s and/or processor/s) 910 ):
The remaining limitations are similar to claim 1. Claim 13 is rejected for the same reasons as claim 1.
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1.
Concerning independent claim 14, Putman and Liu disclose:
A non-transient computer-readable recording medium (Putman - see par 5 - a system having one or more processors, and a non-transitory memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations; see par 96 - Deep learning controller 218 can include a processing unit (e.g., CPU/s and/or processor/s) 910); recording a quality prediction program for a computer to execute (Putman -see par 59 – predict process performance; see par 95 – predictive process control; see par 97 - Processor 910 can be coupled to storage device 930, which can be configured to store software and instructions necessary for implementing one or more functional modules and/or database systems); see par 104 - , non-transitory computer readable media can include media such as non-transitory magnetic media (such as hard disks, floppy disks, etc.), non-transitory optical media (such as compact discs, digital video discs, Blu-ray discs, etc.), non-transitory semiconductor media (such as flash memory, electrically programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), etc.), any suitable media that is not fleeting or devoid of any semblance of permanence during transmission, and/or any suitable tangible media):
The remaining limitations are similar to claim 1. Claim 14 is rejected for the same reasons as claim 1.
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1.
Concerning claim 2, Putman and Liu disclose:
The quality prediction system according to claim 1, further comprising:
a confirmation part configured to confirm whether there is an adjustment item that does not contribute to prediction of the quality index among the each adjustment item by statistically analyzing each sample included in the acquired first learning data and the acquired second learning data (Putman – see par 68 - deep learning controller 218 can consider all parameters of the manufacturing process (e.g., one or more control values, one or more station values, one or more process values, one or more stations, one or more intermediate outputs or any combination thereof), and using one or more of its machine learning algorithms can identify the key influencers. In some aspects, deep learning controller 218 can employ unsupervised machine learning techniques to discover one or more key influencers, for example, wherein each key influencer is associated with one or more parameters (or parameter combinations) that affect characteristics of various station outputs, the final output, and/or process performance. see par 69 - deep learning controller 218 can rank, in order of significance, the impact of each parameter of the manufacturing process on the final output value or the process performance. A key influencer, can be identified based on: a cutoff ranking (e.g., the top 5 aspects of the manufacturing process that impact the final output value), a minimum level of influence (e.g., all aspects of the manufacturing process that contribute at least 25% to the final output value); see par 79 - instead of collecting volumes of data that marginally impact the process, data resources (e.g., collection, processing and storage) can be allocated largely for the data associated with the key influencers (“curated data”). Further, the curated data (a subset of all the data available from the manufacturing process) can be provided to machine learning algorithms to make optimizations to the key influencers, reducing the volume of training examples and increasing the resources available to process the curated data.).
Concerning claim 3, Putman and Liu disclose:
The quality prediction system according to claim 2, wherein the second data acquisition part is further configured to acquire additional second learning data, which is collected by experimenting manufacturing of the second product by the second manufacturing equipment under a plurality of conditions with respect to the adjustment item that does not contribute to prediction of the quality index, in response to determining by the confirmation part that there is the adjustment item that does not contribute to prediction of the quality index (Putman – see par 71 - In some embodiments, as described in connection with FIG. 5, the key influencers can be used to help build a more robust data set to train deep learning controller 218. see par 72 - in some embodiments, deep learning controller 218 can intentionally make changes to control inputs, to create the conditions for generating intermediate output values that may exceed the normal fluctuations of an in control process under traditional manufacturing process control (e.g., SPC), but still generates final outputs that are in specification. This creates a more robust data training set for deep learning controller 218 to detect patterns and determine how particular stations, station/control/process values, and intermediate output values impact the final output value (e.g., whether the final output is in specification or not)), and
the machine learning part is configured to implement machine learning of the quality prediction model further using the acquired additional second learning data (Putman - see par 73 - deep learning controller 218 can adjust known controller inputs (e.g., control setpoints) to one or more station controllers to produce intermediate output values that may exceed a specified range from the mean. For example, once deep learning controller 218 is conditioned (step 335), it knows at least some control inputs for each station controller that will result in final output values that are in specification. In some embodiments, deep learning controller 218 can select one or more station controllers and vary the known control inputs (e.g., setpoints) to the selected station controller(s) by a predetermined threshold (e.g., new setpoint=original setpoint+1% original setpoint) (step 510). Deep controller 218 can predict the expected value (EV) of the final output for the manufacturing process using the newly calculated control inputs, along with a confidence level for its prediction (step 515). If deep learning controller 218 determines, with a threshold confidence level, that the expected value will be in specification (step 517), then deep learning controller 218 can provide the adjusted control inputs to the selected station controllers (step 520). see par 76 - s creating a robust data set by varying a control input (e.g., a temperature setpoint for a particular station). In this example, the setpoint temperature for a specific station is 95°, and the actual temperature of the station fluctuates between 92°-98° (i.e., 3° above and below the setpoint temperature). The setpoint temperature at 95° and the corresponding ±3° fluctuation of the actual station temperature all result in final output values that are in specification.).
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1.
Concerning claim 7, Putman and Liu disclose:
The quality prediction system according to claim 1, wherein the second data acquisition part is further configured to acquire additional second learning data which is further collected in the second manufacturing equipment in response to the quality index of the second product manufactured by the second manufacturing equipment, in which the adjustment amount of the each adjustment item is set to the adjustment amount specified by the specifying part, not satisfying the quality standard (Putman – see par 39 - Unacceptable intermediate output values, which can also be defined by station, refer to upper/lower limits or range of intermediate output values that define the parameters for an intermediate output that will ultimately result in a final output that is not in specification, unless corrective action is taken at another station. see par 66 - , deep learning controller 218 can also monitor whether any of the station/control/process or intermediate output values are unacceptable and make further adjustments to the station controller inputs),
the machine learning part is further configured to implement machine learning of the quality prediction model again further using the acquired additional second learning data (Putman – see par 79 - Once the most important stations, station/control/process values, intermediate output values that influence the final output value or the process performance are identified, then resource allocation and process optimization can target the key influencers. For example, instead of collecting volumes of data that marginally impact the process, data resources (e.g., collection, processing and storage) can be allocated largely for the data associated with the key influencers (“curated data”). Further, the curated data (a subset of all the data available from the manufacturing process) can be provided to machine learning algorithms to make optimizations to the key influencers, reducing the volume of training examples and increasing the resources available to process the curated data. see par 86 - Deep learning controller 218 can also compare the expected values with the actual values to provide feedback on its prediction and to further adjust the control inputs (step 545). As the process begins and goes through all the stations, in some embodiments, only measurements related to the key influencers, as well as measurements for the final output, will be collected and provided to deep learning controller 218. In other embodiments, deep learning controller 218 can continue to collect data from all the process stations. The data will be used to continuously improve the control inputs for the key influencers, so that the optimal design/process values are achieved consistently and with a high confidence level.), and
the specifying part is further configured to specify the adjustment amount of the each adjustment item of the second manufacturing equipment so that the predicted quality index of the second product satisfies the quality standard, using the retrained quality prediction model (Putman – see par 71 - as described in connection with FIG. 5, the key influencers can be used to help build a more robust data set to train deep learning controller 218. see par 72 - As the deep learning controller's predictions become more accurate, in some embodiments, deep learning controller 218 can intentionally make changes to control inputs, to create the conditions for generating intermediate output values that may exceed the normal fluctuations of an in control process under traditional manufacturing process control (e.g., SPC), but still generates final outputs that are in specification. This creates a more robust data training set for deep learning controller 218 to detect patterns and determine how particular stations, station/control/process values, and intermediate output values impact the final output value; see par 74 - control inputs related to material tolerances can be purposely varied, and deep learning controller 218 can be used to determine what adjustments to make to other control inputs to produce in specification final products. By training deep learning controller 218 in this manner, when new materials are introduced into the manufacturing process unexpectedly, deep learning controller 218 can adapt the control inputs on its own, without requiring operator input. ).
Concerning claim 8, Putman and Liu disclose:
The quality prediction system according to claim 7, further comprising a candidate presenting part (Putman see par 45 - one or more process stations can be operated manually, for example, by a human operator performing specific instructions. Instead of an electronic station controller, an operator follows a set of instructions,) configured to specify a condition for each adjustment item in the second manufacturing equipment, which obtains a sample that contributes to improving prediction accuracy of the quality prediction model for the first learning data and the second learning data, according to a predetermined evaluation standard, and present the specified condition (Putman see par 56 - Deep learning controller 218 can provide a confidence level for its prediction at an instant or over a specific time period, for example, to provide a measure of statistical confidence in the prediction. In some aspects, the confidence level may be expressed as a numerical probability of accuracy for the prediction; see par 75 - For example, deep learning controller 218 can adjust the setpoints associated with the key influencers by 1% and iterate through the manufacturing process one or more times. On a subsequent iteration, deep learning controller 218 can adjust the setpoints associated with the key influencers by another 1%, and iterate through the manufacturing process one or more times; see par 87 - once deep learning controller 218 predicts control inputs to achieve optimal design/process value with a certain confidence level over a period of time, deep learning controller 218 can identify which of the key influencers are key influencers for driving the desired optimizations and only target that subset of key influencers, further reducing the possible actions/states that a machine learning algorithm must consider and more efficiently allocating the resources to that subset; see par 88 - deep learning controller 218 can, not only initialize the station controller inputs for the key influencers before the start of an iteration through the process stations of the manufacturing process, but also adjust the control inputs during the process itself. In particular, based on information received from prior stations in a process, deep learning controller 218 can make changes to the control inputs associated with later stations in the process to ensure the optimal design/process values are achieved. ).
Concerning claim 9, Putman and Liu disclose:
The quality prediction system according to claim 1, wherein the first manufacturing equipment is a basic production line that serves as a basis for other production lines (Putman – see par 38 - Functional priors, as used herein, refers to information relating to the functionality and known limitations of each process station, individually and collectively, in a manufacturing process; specifications for the tools/equipment used at the process station are all considered functional priors. see par 39 - Experiential priors, as used herein, refers to information gained by prior experience with, for example performing the same or similar manufacturing process; operating the same or similar stations; producing the same or similar intermediate/final outputs. see par 68 - Deep learning controller 218 can consider all parameters of the manufacturing process (e.g., one or more control values, one or more station values), and
the second manufacturing equipment is a duplicate production line configured by duplicating the basic production line ( Putman – see par 39 - Experiential priors, as used herein, refers to information gained by prior experience with, for example performing the same or similar manufacturing process; operating the same or similar stations; producing the same or similar intermediate/final outputs; see par 55 - Further, the training of deep learning controller 218 can be augmented by: providing deep learning controller 218 with simulated data or data from a similar manufacturing process. see par 72 - more robust data training set for deep learning controller 218 to detect patterns and determine how particular stations, station/control/process values, and intermediate output values impact the final output value (e.g., whether the final output is in specification or not). ).
Concerning claim 10, Putman and Liu disclose:
The quality prediction system according to claim 1, wherein the first manufacturing equipment is a production line before making a state change (Putman – see par 20 - The manufacturing process is complex and comprises raw materials being processed by different process stations (or “stations”) until a final product (referred to herein as “final output”) is produced), and
the second manufacturing equipment is a production line after making the state change (Putman – see par 21 - Each process station can include one or more tools/equipment that performs a set of process steps on: received raw materials (this can apply to a first station or any of the subsequent stations in the manufacturing process) and/or the received output from a prior station (this applies to any of the subsequent stations in the manufacturing process). Examples of process stations can include, but are not limited to conveyor belts, injection molding presses, cutting machines, die stamping machines; see par 83 - deep learning controller 218 can determine the key influencers for driving in specification products (see also FIG. 4), and can predict the control inputs (“optimal control inputs”) to control each key influencer to achieve the desired optimal design or process values).
Concerning claim 11, Putman and Liu disclose:
The quality prediction system according to claim 1, wherein the quality prediction model is configured by a neural network, a support vector machine, or a regression model (Putman – see par 34 - For example, machine learning techniques can utilize one or more of the following, alone or in combination: hidden Markov models; recurrent neural networks; convolutional neural networks (CNNs); deep learning; Bayesian symbolic methods; reinforcement learning, general adversarial networks (GANs); support vector machines).
Claims 4 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Putman (US 2020/0401119) in view of Liu et al., "Domain adaptation transfer learning soft sensor for product quality prediction," 2019, Chemometrics and Intelligent Laboratory Systems, Vol. 192, No. 103813, pages 1-11, as applied to claims 1-3 and 7-14 above, and further in view of Sharma (US 2017/0293269).
Concerning claim 4 and 16, Putman and Liu disclose:
The quality prediction system according to claim 2, wherein statistically analyzing the each sample comprises:
standardizing a value of each adjustment item of each sample included in the first learning data and the second learning data (Putman – see par 71 - in connection with FIG. 5, the key influencers can be used to help build a more robust data set to train deep learning controller 218. see par 75 - deep learning controller 218 can first determine the key influencers (step 446 as described in connection with FIG. 4) and vary the control inputs associated with the key influencers. To create a robust data set, deep learning controller 218 can continue to adjust the control inputs associated with specific station controllers each time it iterates through the manufacturing process (step 528). For example, deep learning controller 218 can adjust the setpoints associated with the key influencers by 1% and iterate through the manufacturing process one or more times. On a subsequent iteration, deep learning controller 218 can adjust the setpoints associated with the key influencers by another 1%, and iterate through the manufacturing process one or more times. );
calculating … each adjustment item from the standardized value of the each sample (Putman –see par 36 - Machine learning models, as discussed herein, can also be used to determine the process stations, control/station/process values and intermediate output values that are most influential on the final output value (“key influencers”), and to optimize the manufacturing process by targeting the key influencers. see par 56 - Deep learning controller 218 can provide a confidence level for its prediction at an instant or over a specific time period, for example, to provide a measure of statistical confidence in the prediction. In some aspects, the confidence level may be expressed as a numerical probability of accuracy for the prediction, in other aspects, the confidence level may be expressed as an interval or probability range; see par 75 - , deep learning controller 218 can continue to adjust the control inputs associated with specific station controllers each time it iterates through the manufacturing process (step 528). )
Sharma discloses:
calculating “variance” of the each adjustment item from the standardized value of the each sample (Sharma – see par 30 - The input data set for which predictions are to be made by the model may include real-time sensor data that includes measurements of monitored manufacturing processes. If predictions generated by the model estimate that a process or device being manufactured will fall outside tolerances, alerts may be generated and corrective actions may be implemented by the systems and devices 120 to account for the out-of-tolerance predictions. See par 42 - According to an example, the model 150 is a structural time series model, and includes a time-varying parameter (β), an independent time-varying variable (X) and a dependent time-varying variable (Y). The model 150 may include multiple independent variables X.sub.j that each contribute to the value of Y. The model 150 determines the effectiveness of the independent variable X to contribute to the dependent variable Y for a given time period t, and β is associated with the effectiveness; see par 56 - As discussed in the method 400, a variance in β is computed to determine if the variance is greater than a threshold. If so, steps may be performed for determining whether a change to at least one of the β's is warranted. Relative Standard Deviation (RSD), which is a modified standard deviation, may be used to measure the variance of the β's. The RSD is defined as standard deviation normalized by its mean (or its absolute value). RSD may be used to determine whether a change to at least one of the β's is warranted)
Putman, Liu, and Sharma disclose:
recognizing an adjustment item with a calculated value of variance being 0 as the adjustment item that does not contribute to prediction of the quality index (Putman – see par 69 - deep learning controller 218 can rank, in order of significance, the impact of each parameter of the manufacturing process on the final output value or the process performance. A key influencer, can be identified based on: a cutoff ranking (e.g., the top 5 aspects of the manufacturing process that impact the final output value), a minimum level of influence (e.g., all aspects of the manufacturing process that contribute at least 25% to the final output value); or any other suitable criteria. see par 79 - For example, instead of collecting volumes of data that marginally impact the process, data resources (e.g., collection, processing and storage) can be allocated largely for the data associated with the key influencers (“curated data”). Further, the curated data (a subset of all the data available from the manufacturing process) can be provided to machine learning algorithms to make optimizations to the key influencers, reducing the volume of training examples and increasing the resources available to process the curated data. In addition, the machine learning algorithms can be directed to optimizing the key influencers; see par 77 - once deep learning controller 218 predicts control inputs to achieve optimal design/process value with a certain confidence level over a period of time, deep learning controller 218 can identify which of the key influencers are key influencers for driving the desired optimizations and only target that subset of key influencers, further reducing the possible actions/states that a machine learning algorithm must consider and more efficiently allocating the resources to that subset;
Sharma see par 56 - As discussed in the method 400, a variance in β is computed to determine if the variance is greater than a threshold. If so, steps may be performed for determining whether a change to at least one of the β's is warranted. Relative Standard Deviation (RSD), which is a modified standard deviation, may be used to measure the variance of the β's. The RSD is defined as standard deviation normalized by its mean (or its absolute value). RSD may be used to determine whether a change to at least one of the β's is warranted).
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1. In addition, Putman, Liu, and Sharma are analogous art as they are directed to predicting faults/quality for manufacturing output/products (See Putman Abstract; Liu Abstract, par 57; Sharma Abstract, par 30, 62 – forecasting of manufacturing conforming to specifications). Putman discloses determine the “most influential” process values that are key influencers for a manufacturing process (See par 36) and using deep learning to measure confidence level in prediction (See par 56) and only targeting “key influencers” for optimization (See par 77). Sharma improves upon Putman and Liu by disclosing including variables that contribute to an output value along with effectiveness to contribute (See par 42) and measuring standard deviation and variance to determine whether a change is warranted (See par 56). One of ordinary skill in the art would be motivated to further include using variance for which variables to possibly change that effect an output variable to efficiently improve upon the “most influential process values being key influencers for a manufacturing process in Putman (See par 55, 91) and the domain adaptation and transfer learning in Liu.
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system and method of using machine learning for adjusting control inputs if expected values are not in-specification in Putman (Abstract, par 63) to further use transfer learning for different types/grades of manufacture from different domains for quality prediction as disclosed in Liu, and further including consideration for standard deviation and variance for variables that warrant changing for contributing to an output value as disclosed in Sharma, since the claimed invention is merely a combination of old elements, and in combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable and there is a reasonable expectation of success.
Claims 5 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Putman (US 2020/0401119) in view of Liu et al., "Domain adaptation transfer learning soft sensor for product quality prediction," 2019, Chemometrics and Intelligent Laboratory Systems, Vol. 192, No. 103813, pages 1-11, as applied to claims 1-3 and 7-14 above, and further in view of Kruger, et al., "Intelligent machine agent architecture for adaptive control optimization of manufacturing processes," 2011, Advanced Engineering Informatics, Vol. 25, no. 4, pages 783-796.
Concerning claim 5 and 17, Putman discloses determining control/station/process values most influence on final output value, as in “key influencers” to optimize manufacturing process, using machine learning (See par 36) and analyzing key influencers by cutoff or minimum level (See par 69).
Kruger discloses:
The quality prediction system according to claim 2, wherein statistically analyzing the each sample comprises:
principal-component analyzing each sample included in the first learning data and the second learning data (Kruger see page 790, section 4.2.2 - Data features representing quantities influenced by the goal variables must be identified and/or combined to reduce the dimensionality of the data and provide a signal highly correlated to the goal variable, in order to optimize the model building process
[49]. Multiple variables may measure the same driving principle governing the behavior of the process. In this case, a group of variables exhibiting similar variance can be grouped and replaced with a single new variable [50]. There are many linear and non-linear techniques to achieve this. Probably, the best known linear technique is Principle Component Analysis (PCA), however non-linear
variations of PCA also exist (i.e. kernel PCA see page 794, col. 1, 2nd paragraph - Following feature extraction, dimensional reduction was performed
using Principal Component Analysis (PCA) to reduce the dimensionality of the data and to remove noise from the extracted sensor data features. see par 791, col. 1 - Fig. 5 illustrates the PCA process on a set of sensor data. It can be seen that the 20 possible sensor and transformation combinations have been grouped into two variables (PC1 and PC2),).
Putman and Kruger (for “principal component”) disclose:
recognizing an adjustment item corresponding to a principal component whose contribution rate is lower than a threshold value among principal components obtained by principal component analysis as the adjustment item that does not contribute to prediction of the quality index (Putman – see par 36 - Machine learning models, as discussed herein, can also be used to determine the process stations, control/station/process values and intermediate output values that are most influential on the final output value (“key influencers”), and to optimize the manufacturing process by targeting the key influencers. see par 63 - Further, if deep learning controller 218 determines that the final output is not in-specification, it can calculate adjustments to the control inputs, so that the predicted expected value for the final output is in-specification.
Kruger – see page 792, col. 1, last paragraph - The process parameters, PCs, goal metrics, process states matrix obtained from this iteration of the perception phase are stored in the episodic memory along with the data from previous iterations, forming a set of experiences the agent has perceived in its environment. The goal of feature selection is to establish an inference chain between the process parameters (technical) and process performance (economic) to obtain the optimal matrix, M. The optimal matrix will be passed to the modeling phase to convert this relationship into a predictive model, allowing the agent to learn about the process. see page 794, col. 2, 2nd paragraph - Feature selection, using the Filter method, was used to select the optimal data to develop the PU-UG/US models. can be seen that all features from PC1 and PC2 are significant, showing that the sensor data features contribute to both PCs and that both groups of data are related to drill tool wear. This was expected based on the results of the PCA. This may not always be the case and is dependent on the process. overall best combination of the two correlations is selected as the optimal PUUG/ US chain. Ultimately, PC 2 feature 2 was selected as both spindle
speed and federate were found to be significant contributors (p < 0.001) to this feature and PC. The agent has therefore established the inference chain from the process parameters to the process performance metric utilizing the sensor data to provide further insight into this relationship.).
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1. In addition, Putman, Liu, and Kruger are analogous art as they are directed to predicting faults/quality for manufacturing output/products (See Putman Abstract; Liu Abstract, par 57; Kruger Abstract, page 783, col. 2, 2nd paragraph factors include “quality”; page 787 – final quality depends on processes utilized during production). Putman discloses determining control/station/process values most influence on final output value, as in “key influencers” to optimize manufacturing process, using machine learning (See par 36) and analyzing key influencers by cutoff or minimum level (See par 69). Kruger improves upon Putman and Liu by disclosing use of principal component analysis for which features to adjust that contribute to a prediction of a process performance metric. One of ordinary skill in the art would be motivated to further include using principal component analysis for contributing to adjustments for improving production metrics to efficiently improve upon the “most influential process values being key influencers for a manufacturing process in Putman (See par 55, 91) and the domain adaptation and transfer learning in Liu.
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system and method of using machine learning for adjusting control inputs if expected values are not in-specification in Putman (Abstract, par 63) to further use transfer learning for different types/grades of manufacture from different domains for quality prediction as disclosed in Liu, and further including principal component analysis for which features to adjust that contribute to a prediction of a process performance metric as disclosed in Kruger, since the claimed invention is merely a combination of old elements, and in combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable and there is a reasonable expectation of success.
Claims 6 and 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Putman (US 2020/0401119) in view of Liu et al., "Domain adaptation transfer learning soft sensor for product quality prediction," 2019, Chemometrics and Intelligent Laboratory Systems, Vol. 192, No. 103813, pages 1-11, as applied to claims 1-3 and 7-14 above, and further in view of Applicant’s Admitted Prior Art (AAPA) of Daume, “Frustratingly Easy Domain Adaptation”, 2007, in Proceedings of the 45th, Annual Meeting of the Association of Computational Linguistics, pp. 256-263.
Concerning claims 6, 18, and 19, Liu discloses that there is a “simple” transfer learning method for domain adaptation (See page 2, 2nd paragraph).
The quality prediction system according to claim 1, wherein the data transfer part is configured to convert the acquired first learning data to match the second learning data by a Frustratingly Easy Domain Adaptation method or a Correlation Alignment method (Liu discloses that there is a “simple” transfer learning method for domain adaptation (See page 2, 2nd paragraph).).
To the extent “Frustratingly Easy Domain Adaptation” is a different feature, Applicant’s Admitted Prior Art of Daume is applied:
The quality prediction system according to claim 1, wherein the data transfer part is configured to convert the acquired first learning data to match the second learning data by a Frustratingly Easy Domain Adaptation method or a Correlation Alignment method (Applicant’s Admitted Prior Art in [0089] as published states “FIG. 7 schematically shows an example of the method of converting the first learning data 30 and the second learning data 35 by the Frustratingly Easy Domain Adaptation method. The Frustratingly Easy Domain Adaptation method is the method disclosed in Reference 1 “H. Daume III (University of Utah), ‘Frustratingly Easy Domain Adaptation’, in Proceedings of the 45tt, Annual Meeting of the Association of Computational Linguistics, pp. 256-263 (2007))”.
See Daume title of “Frustratingly Easy Domain Adaptation,” page 256, Abstract “our approach is incredibly simple, easy to implement”; page 256, Col. 2, 2nd paragraph “One particularly nice property of our approach is that it is incredibly easy to implement: the Appendix provides a 10 line, 194 character Perl script for performing the complete transformation”).
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1. In addition, Putman, Liu, and Applicant’s Admitted Prior Art (AAPA) of Daume are analogous art as they are concerned with learning from other data sets (Putman – par 39 – experiential priors from similar manufacturing process or similar stations; Liu Abstract, par 57; Daume Abstract, page 256). Liu discloses that there is a “simple” transfer learning method for domain adaptation (See page 2, 2nd paragraph). AAPA of Daume improves upon Putman and Liu by disclosing use of “frustratingly easy domain adaptation”. One of ordinary skill in the art would be motivated to further include using the known “frustratingly easy domain adaptation” to efficiently improve upon the “most influential process values being key influencers for a manufacturing process in Putman (See par 55, 91) and the domain adaptation and “simple” transfer learning in Liu.
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system and method of using machine learning for adjusting control inputs if expected values are not in-specification in Putman (Abstract, par 63) to further use transfer learning for different types/grades of manufacture from different domains for quality prediction as disclosed in Liu, and further use “Frustratingly Easy Domain Adaptation” as disclosed in AAPA of Daume, since the claimed invention is merely a combination of old elements, and in combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable and there is a reasonable expectation of success.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Putman (US 2020/0401119) in view of Liu et al., "Domain adaptation transfer learning soft sensor for product quality prediction," 2019, Chemometrics and Intelligent Laboratory Systems, Vol. 192, No. 103813, pages 1-11, and further in view of Azamfar, "Deep learning-based domain adaptation method for fault diagnosis in semiconductor manufacturing," 2020, IEEE Transactions on Semiconductor Manufacturing Vol. 33, no. 3, pages 445-453.
Concerning independent claim 15, Putman and Liu disclose:
A quality prediction system (Putman – see par 5 - a system having one or more processors; see par 21 - Each process station can include one or more tools/equipment that performs a set of process steps on: received raw materials (this can apply to a first station or any of the subsequent stations in the manufacturing process) and/or the received output from a prior station (this applies to any of the subsequent stations in the manufacturing process). Examples of process stations can include… quality control), comprising:
The next limitations are similar to claim 1. Claim 15 is rejected for the same reasons as claim 1.
Liu discloses having an output matrix that is a weighted combination between a source and target domain (See page 4).
Azamfar discloses the remaining limitations:
wherein the quality prediction model comprises a kernel function (Azamfar – see page 447, col. 2, Section B - Transfer learning has been popularly developed as a promising technique to address the practical problems where the training and testing data are from different distributions. the popular maximum mean discrepancy (MMD) metric is used to measure and optimize the distribution
discrepancy between the source and target domains in domain adaptation. Basically, the MMD metric is defined as the squared distance between the kernel embeddings of data marginal distributions in the reproducing kernel Hilbert space (RKHS) as), and
the machine learning comprises constructing a kernel function for the first learning data by weighting a kernel function for the second learning data, and adjusting a value of each parameter of the quality prediction model based on statistical criteria (Azamfar – see page 448, col. 1, 1st paragraph – kernel choice is very important to enhance the performance of domain adaptation. That is because different kernels can embed the probability distributions in different RKHSs where different orders of the concerned statistics are focused on. Hence, multiple kernels in MMD are used in this study to take advantage of different kernels and obtains a principled method for kernel selection. where kσi denotes a Gaussian kernel with bandwidth coefficient σi.).
It would have been obvious to combine Putman and Liu for the same reasons as discussed with regards to claim 1. In addition, Putman, Liu, and Azamfar are analogous art as they are directed to predicting faults/quality for manufacturing output/products (See Putman Abstract; Liu Abstract, par 57; Azamfar abstract, page 445, col. 1, 1st paragraph – predicting manufacturing quality) and where both Liu and Azamfar are concerned with “Transfer Learning” (Liu Abstract, Azamfar page 447, col. 2). Putman mentions in [0091] that its system optimization can reduce possible actions/states by incorporating by reference “62/836,213 “Transfer Learning Approach to Multi-Component Manufacturing Control”. Liu discloses having an output matrix that is a weighted combination between a source and target domain (See page 4). Azamfar improves upon Putman and Liu by disclosing using kernels from different probabilities and coefficients when adapting from one domain to another. One of ordinary skill in the art would be motivated to further include using “using kernels from different probabilities and coefficients when adapting from one domain to another to efficiently improve upon the use of data from “similar” manufacturing, where “Transfer Learning” is mentioned in Putman (See par 55, 91) and the domain adaptation and transfer learning in Liu.
Accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system and method of using machine learning for adjusting control inputs if expected values are not in-specification in Putman (Abstract, par 63) to further use transfer learning for different types/grades of manufacture from different domains for quality prediction as disclosed in Liu, and further including kernel functions from different probabilities and coefficients as disclosed in Azamfar, since the claimed invention is merely a combination of old elements, and in combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable and there is a reasonable expectation of success.
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Putman (US 2020/0401119) in view of Liu et al., "Domain adaptation transfer learning soft sensor for product quality prediction," 2019, Chemometrics and Intelligent Laboratory Systems, Vol. 192, No. 103813, pages 1-11, and Sharma (US 2017/0293269), as applied to claim 4 and 16 above, and further in view of Applicant’s Admitted Prior Art (AAPA) of Daume, “Frustratingly Easy Domain Adaptation”, 2007, in Proceedings of the 45th, Annual Meeting of the Association of Computational Linguistics, pp. 256-263.
Concerning claim 20, Liu discloses that there is a “simple” transfer learning method for domain adaptation (See page 2, 2nd paragraph).
The quality prediction system according to claim 4, wherein the data transfer part is configured to convert the acquired first learning data to match the second learning data by a Frustratingly Easy Domain Adaptation method or a Correlation Alignment method (Liu discloses that there is a “simple” transfer learning method for domain adaptation (See page 2, 2nd paragraph)).
To the extent “Frustratingly Easy Domain Adaptation” is a different feature, Applicant’s Admitted Prior Art of Daume is applied:
The quality prediction system according to claim 4, wherein the data transfer part is configured to convert the acquired first learning data to match the second learning data by a Frustratingly Easy Domain Adaptation method or a Correlation Alignment method (Applicant’s Admitted Prior Art in [0089] as published states “FIG. 7 schematically shows an example of the method of converting the first learning data 30 and the second learning data 35 by the Frustratingly Easy Domain Adaptation method. The Frustratingly Easy Domain Adaptation method is the method disclosed in Reference 1 “H. Daume III (University of Utah), ‘Frustratingly Easy Domain Adaptation’, in Proceedings of the 45tt, Annual Meeting of the Association of Computational Linguistics, pp. 256-263 (2007))”.
See Daume title of “Frustratingly Easy Domain Adaptation,” page 256, Abstract “our approach is incredibly simple, easy to implement”; page 256, Col. 2, 2nd paragraph “One particularly nice property of our approach is that it is incredibly easy to implement: the Appendix provides a 10 line, 194 character Perl script for performing the complete transformation”).
It would have been obvious to combine Putman, Liu, Sharma, and AAPA of Daume for the same reasons as discussed with regards to claim 1, 4, and 6. Liu discloses that there is a “simple” transfer learning method for domain adaptation (See page 2, 2nd paragraph). AAPA of Daume improves upon Putman and Liu and Sharma by disclosing use of “frustratingly easy domain adaptation”. One of ordinary skill in the art would be motivated to further include using the known “frustratingly easy domain adaptation” to efficiently improve upon the “most influential process values being key influencers for a manufacturing process in Putman (See par 55, 91) and the domain adaptation and “simple” transfer learning in Liu.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure: Kawanoue (US 2019/0286085) – directed to abnormality detection using a model with features, where “state values” include position, speed, torque (Abstract, par 35).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to IVAN R GOLDBERG whose telephone number is (571)270-7949. The examiner can normally be reached 830AM - 430PM.
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, Anita Coupe can be reached at 571-270-3614. 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.
/IVAN R GOLDBERG/ Primary Examiner, Art Unit 3619
Prosecution Insights