US 11402438

I'll search for authoritative information on US 11,402,438, including any 2026 CAFC docket activity.

Let me do targeted searches for any litigation or CAFC activity involving this patent number or its owners.

US Patent 11,402,438 — Summary

Note on scope of search: I searched for this exact number (11402438 / 11,402,438) plus assignee names against public patent databases and Federal Circuit / district-court document results. I found no CAFC 2026 docket activity, IPR, or district-court litigation naming this patent. I cannot rule out a pending matter that is not indexed in the sources searched — treat the "no litigation found" statement as a negative search result, not a definitive clearance.

Bibliographic data

Abstract (as issued)

Disclosed is an intelligent on-line diagnosis method for winding deformation of power transformers. Windings may undergo local twisting, swelling or the like under electric power or mechanical force (winding deformation), a hidden danger to safe network operation. Conventional diagnosis methods are all off-line, requiring shutdown and highly skilled operators. The invention provides an intelligent on-line diagnosis method based on a combination of information entropy and support vector machine: feature extraction of current and voltage signals based on permutation entropy and wavelet entropy, integrating variation of monitoring indicators in complexity and time-frequency domain, and automatically learning diagnostic logic from fault features via machine learning.

Claim structure

There is one independent claim (claim 1); claims 2–7 are dependent. All seven claims are method claims.

Claim 1 (independent) — plain language:
A method for diagnosing and locating winding deformation in power transformers, in five steps:

1. Step 1: Take the on-line monitoring indicators of each of n known transformers whose deformation positions are known, and split them, following a three-phase/three-winding arrangement, into 9 "position subsamples" (one per winding-phase position) to serve as modeling samples for position diagnosis.
2. Step 2: For the 9n subsamples, compute root-mean-square errors of (a) normalized permutation entropy, (b) wavelet entropy, and (c) an average, of the monitored indicators across two sequences — before and after the last short circuit — to build the feature set.
3. Step 3: Attach a tag indicating whether deformation occurred to each subsample's feature set and feed the tagged sets into an SVM for classification learning, producing a trained SVM.
4. Step 4: Run hierarchical cross-validation to measure accuracy, precision and recall.
5. Step 5: For a transformer under test, first run the on-line deformation diagnosis method to confirm deformation, then split its indicators into 9 position subsamples the same way, compute the feature set as in Step 2, feed it into the trained SVM, and output a per-subsample diagnosis result (i.e., which winding position is deformed).

Dependent claims (brief):

• Claim 2 — defines the 9 positions: HV A/B/C, MV A/B/C, LV A/B/C.
• Claim 3 — defines the indicator set: voltage and current monitoring data plus 12 new phase-difference indicators constructed from a three-phase unbalance rate, 30 indicators total; gives the LV-side formulas (e.g., current difference between LV phases A and B = LV phase B current amplitude − LV phase A current amplitude), applied analogously at MV and HV.
• Claim 4 — details the permutation entropy algorithm: phase-space reconstruction by delay coordinates into an m × K matrix (m = embedding dimension, τ = delay time); rank-ordering of each reconstructed vector; counting of the at-most m! arrangement patterns; relative frequency P(i); entropy H = −ΣP(i)·ln P(i); and normalization by ln(m!).
• Claim 5 — details the wavelet entropy algorithm: multi-level filtering into high-frequency detail sub-bands D1…DM and a low-frequency approximation sub-band AM(n); band energies E1…En; and total energy entropy H = −Σ e(i)·ln e(i) using relative band energies.
• Claim 6 — defines the Step-2 metrics as RMSE_PE, RMSE_WE and RMSE_AVG, and describes computing the root of the mean of squared entropy differences across all monitoring indicators for the pre-/post-short-circuit transform.
• Claim 7 — sets out sub-steps (a)–(f) of the Step-5 on-line diagnosis method: (a) build current and voltage phase-difference indicators by subtraction across phases; (b) split data into two sequences — before/after the last short circuit, or first half/second half if no short circuit occurred; (c) compute the RMSE entropy/average features; (d) tag and train the SVM; (e) hierarchical cross-validation; (f) apply the same feature extraction to the test sample and output the SVM diagnosis.

Technical substance

The patent's contribution is combining information-entropy features (permutation entropy for complexity/ordinal structure; wavelet entropy for time-frequency energy distribution) with an SVM classifier to move winding-deformation diagnosis from off-line (frequency response, low-voltage short-circuit impedance, dielectric-loss capacitance tests requiring outage) to on-line using existing voltage/current monitoring data. The "positioning" aspect comes from decomposing the transformer into 9 winding-position subsamples and classifying each separately, so the output localizes the deformation rather than merely flagging it. The specification reports an application example with 29 modeling transformers; for position diagnosis only three transformers (HB, WT, XX) had known deformed/non-deformed positions, giving 27 position subsamples, with a 2-D scatter plot (FIG. 4) showing deformed subsamples clustering in the upper-right (larger pre/post difference) versus normal subsamples in the lower-left.

Uncertainties

• No authoritative source indicates any US litigation, CAFC appeal, or PTAB proceeding for US 11,402,438. My searches returned unrelated matters that merely mention other "438" patents (e.g., U.S. 8,441,438 in CyWee v. Samsung), which must not be confused with this patent. Verify against PACER/Docket Navigator and the CAFC docket if litigation status is material.
• The patent text renders "SVM" once as "Small Vector Machine" (a translation artifact); the correct expansion used throughout is Support Vector Machine.
• Claim 1's Step 5 depends on a prior on-line deformation diagnosis result ("first carrying out an on-line winding deformation diagnosis method to conclude that the transformer to be tested is deformed") — this two-stage gating (detect, then localize) is worth noting when assessing claim scope.

I'll search litigation databases for any cases involving US patent 11,402,438.

Initial searches returned only unrelated patents (e.g., the '438 patent in CyWee v. Samsung is U.S. 8,441,438 — a different patent). Let me run more targeted queries.

No litigation hits yet. Let me check dedicated litigation databases directly.

Bottom line

No litigation involving U.S. Patent No. 11,402,438 ("the '438 patent") was found. I could not identify any district court case, Federal Circuit appeal, ITC investigation, PTAB post-grant proceeding, or foreign (e.g., Chinese) court action in which U.S. 11,402,438 was asserted, challenged, or otherwise at issue.

Because patent litigation data is not something I can confirm exhaustively from open web search alone, the correct statement is: no known litigation; none found. This is a "did not find," not a verified "does not exist."

Patent identification (confirmed)

• Patent: U.S. 11,402,438 B2 — "Intelligent on-line diagnosis and positioning method for winding deformation of power transformers"
• Application: 16/553,665; filed 2019-08-28; granted 2022-08-02
• Priority: CN 201910091394.9, filed 2019-01-30 (published also as CN109884459A / CN109884459B)
• Assignees: Hangzhou Kelin Electric Co., Ltd. and State Grid Zhejiang Electric Power Company Limited Electric Power Research Institute
• Pre-grant publication: US 2020/0241066 A1
• Status: Active; adjusted expiration listed as 2040-12-11

Both assignees are Chinese entities (a Chinese equipment supplier and a State Grid provincial research institute).** That context matters: utilities and their captive research institutes rarely assert patents offensively, which is consistent with the absence of litigation.

What I searched and what I found

I was unable to search:

• PACER — requires a paid, authenticated account; I cannot log in.
• Docket Alarm / Lex Machina / RPX / Docket Navigator — subscription-gated; only fragments surfaced in third-party filings.
• Unified Patents litigation caselist — the specific case-list page did not surface in results for this number.
• Chinese court dockets (e.g., 中国裁判文书网) — not indexed in the results returned.

Important false positives — do not conflate these

While searching, several different patents with superficially similar identifiers appeared. Per your instruction to interpret identifiers literally, I am flagging them explicitly so they are not misattributed:

1. U.S. 8,441,438 — asserted in CyWee Group Ltd. v. Samsung Electronics Co. (E.D. Tex. No. 2:17-cv-00140-WCB-RSP) and in the Google and ZTE IPRs. This is an 8,441,438 patent, not 11,402,438. Unrelated.
2. EP 3082130 A1 — its reference list contains the entries "US 11402437 B" and "US 11402438 B." These are not citations of U.S. Patent Nos. 11,402,437/11,402,438. They are artifacts of the specification's text referring to U.S. application serial Nos. 11/402,737, 11/402,437 and 11/402,438, which the EP reference-generating software mangled into "US 1140243x B" format. No relationship to the '438 patent at issue.
3. U.S. 11,402,438 appeared in the Chinese-language enforcement table (10jqka.com.cn) — that "11402438" is an internal record ID for a 2018 administrative penalty against a clinic, not a patent. Unrelated.

Caveats and recommended verification steps

• Absence of evidence in open web sources is not proof of absence. A patent that has never been litigated simply generates no docket, so there is usually nothing to find — which is the most likely explanation here.
• A definitive negative requires a subscription litigation database query keyed to the patent number across all districts, plus PTAB (IPR/PGR/CBM), the ITC, and the Federal Circuit. I searched for PTAB activity and found none, but I could not run a complete database query.
• To close this out definitively, I recommend running the number 11,402,438 in: Docket Navigator or Lex Machina (litigation), the PTAB's Patent Trial Appeal Board End-to-End (post-grant), and — given the Chinese origin and assignees — 中国裁判文书网 (China Judgments Online) and 国家知识产权局 (CNIPA) invalidation records, since any enforcement would more plausibly occur against Chinese transformer manufacturers in Chinese courts, and would not appear in U.S. dockets at all.

If you can provide authenticated access to PACER or a litigation database export, I can confirm whether the negative holds.

I'll verify the structured "no PTAB proceedings" default against live sources before writing.

Let me run a few more targeted checks before I commit to the negative finding.

PTAB / AIA Trial Analysis — U.S. Patent No. 11,402,438

Proceedings overview

Total AIA trial proceedings on file: 0 — zero IPRs, zero PGRs, zero CBMs, zero DERs (active: 0; claims invalidated: 0; claims sustained: 0; settled: 0; institution denied: 0). The canonical USPTO ODP "PTAB proceedings on file" block returns no proceedings, and my independent web checks surfaced none; the bottom line for a defendant is therefore not "the patent has been hardened by surviving IPRs," but rather "the patent is PTAB-virgin" — no claim of U.S. 11,402,438 has ever been construed, tested, or canceled in an AIA trial, and there is no PTAB estoppel, no adverse FWD reasoning, and no Board claim construction for you to borrow. The entire validity fight, if any, is still ahead of you.

Proceeding inventory

No per-proceeding sections are possible, because there are no proceedings to describe. I do not invent docket numbers. Instead, here is the evidence base for the negative:

Source for the legal-events timeline: https://patents.google.com/patent/US11402438/en

Why the zero is credible here (not just "the search missed it")

• Prosecution was short and narrow: one non-final Office action mailed 2022-02-09, response entered 2022-03-05, notice of allowance 2022-05-13, issue fee 2022-06-14, grant 2022-08-02. A single-rejection, single-response allowance means the examiner did not develop a substantive § 102/§ 103 record against this art. I have not read the substance of that response or any amendment; the file wrapper may contain a narrowing amendment worth obtaining, but I cannot characterize it here.
• Both assignees are Chinese entities — an equipment supplier (Hangzhou Kelin Electric Co., Ltd.) and a State Grid provincial research institute (Electric Power Research Institute of State Grid Zhejiang Electric Power Co., Ltd.). Per the earlier litigation section of this analysis, no U.S. litigation was found in which this patent is asserted. No assertion → no defendant → no § 315(b) clock → no IPR petitioner. The absence of PTAB activity is fully explained by the absence of assertion.
• The PGR window closed unexercised. Because the patent granted 2022-08-02, the § 321(c) nine-month PGR window ran out on 2023-05-02. PGR is permanently foreclosed for this patent.
• CBM was never available in practice. The claims are directed to a technical measurement/diagnostic method classified in G01R 31/62 (testing transformers), not to a "financial product or service" under § 18(d)(1) of the AIA; in any event the CBM program sunset for institution on 2020-09-16, before this patent even issued.

Strategic summary

Claim status: 7 of 7 claims UNTESTED; 0 canceled, 0 sustained. Claims 1–7 all stand exactly as granted. There is no narrowing certificate, no statutory disclaimer on the face of the record, and no adverse judgment. If you are looking for "the claim that killed this patent," there isn't one — the earlier sections of this analysis correctly identify claim 1 (the five-step entropy-plus-SVM method) as the sole independent claim, with claims 2–7 as dependents, and all seven remain live for § 282 purposes.

Estoppel landscape: none exists, which is good news. § 315(e)(2) estoppel attaches only to a petitioner (and its real parties in interest and privies) that obtains an FWD. There is no petitioner, so no one is estopped, no ground is "reasonably could have raised"-barred, and no § 325(e) bar applies to any third party. Practically, this means the full universe of § 102/§ 103 art is still available to a first challenger, both at the PTAB and in district court, subject only to the ordinary § 315(b) one-year bar (triggered only if you are served with a complaint alleging infringement) and the § 315(a)(1) bar (triggered if you file a declaratory-judgment action of invalidity first — file the IPR before the DJ, or the DJ will cause your IPR to be dismissed). Note also the reciprocal risk: because no ground has been raised, a future petitioner gets a clean first-mover position with no claim-construction guidance against it.

Pattern signals: none — no repeat petitioner, no patent-owner PTAB appeals, no defensive aggregator. There is no history of the patent owner aggressively appealing Board decisions (there are no Board decisions), and no indication of a Unified Patents or similar aggregator petition in the chain. The prior-art citations of record are five pre-grant publications, all cited in the original IDS, including US 2020/0200813 A1 (Zhejiang University, "Online diagnosis method for deformation position on trasnformation winding," effective filing 2018-12-21) — an earlier-filed, same-field application publication that predates this patent's 2019-01-30 priority date and is a plausible § 102(a)(2) / § 103 reference, already on the face of the patent. Its being of record reduces its "new art" value but does not reduce its substantive reach; a challenger should start there because it demonstrates the examiner was aware of the closely analogous Chinese-origin winding-deformation-position art.

One policy context worth flagging (hedged): IPR institution practice has been tightening. Unified Patents' amicus brief in Supreme Court No. 25-1230 (filed 2026-05-29) argues the PTO's 2025 revisions — a "settled expectations" discretionary-denial doctrine invoked in hundreds of denials, heightened stipulation requirements for parallel-litigation parties, and a narrowed "compelling merits" exception — have materially restricted institution. That document is a party brief, not an authoritative statement of current Board practice, and this patent (issued 2022-08-02) is not obviously an "older patent" within that doctrine. But the practical takeaway is real: if you intend to file against this patent, expect a higher discretionary-denial risk than the historical ~60–70% institution norm, and consider whether you have a genuine parallel-litigation trigger that would now require a Sargent-type stipulation. Source: https://www.supremecourt.gov/DocketPDF/25/25-1230/[412129](/patent/412129)/20260529145347123_Unified%20Amicus%20Brief%2025-1230.pdf

Recommended next steps

1. You have no PTAB short-cuts available, so this is a merits case, not a leverage case. There is no FWD to link to, no canceled claim to point at, and no Board reasoning to quote. Any statement to your adversary or to a court that a PTAB tribunal has invalidated any claim of U.S. 11,402,438 would be false. Drop that argument entirely.
2. IPR remains the only AIA vehicle — file it, and file it first. The § 321(c) PGR window closed 2023-05-02, and CBM is unavailable as a matter of program sunset and subject-matter scope. IPR under § 311(b) (patents and printed publications, § 102/§ 103 only) is available now and throughout the remainder of the term, which runs to the adjusted expiration of 2040-12-11. Watch § 315(b): if you have been served with a complaint alleging infringement of this patent more than one year ago, you are barred. If you have not yet been served and want to preserve the option, do not file a DJ of invalidity first (§ 315(a)(1)).
3. Build the ground on the art already in the patent's own citation list, then add non-patent literature. Start with US 2020/0200813 A1 (Zhejiang University) as a § 102(a)(2)-based ground, and pair it with the Chinese-language literature in this field (CNKI-indexed transformer winding-deformation and permutation-entropy papers) as printed publications under § 311(b). The claimed entropy formulations in claims 4–6 (normalized permutation entropy, wavelet energy entropy) are textbook material and may be attacked as conventional algorithm elements in a § 103 combination.
4. Attack claim 1 on its two-stage gating, not just its math. Because claim 1 requires, as a preliminarily performed step, "first carrying out an on-line winding deformation diagnosis method to conclude that the transformer to be tested is deformed," a § 112 and § 103 theory can both exploit the functional/open-ended nature of that step. Separately, verify the file wrapper for the 2022-03-05 response to the single 2022-02-09 non-final rejection — I have not read it, and any amendment or argument there is the most likely source of either a § 112 written-description problem or a prosecution-history narrowing you can use.
5. Check the Chinese side before you invest. The same invention issued as CN 109884459 B (granted 2020-12-01; published as CN 109884459 A on 2019-06-14). CNIPA invalidation practice differs materially from the PTAB, and I did not complete a search of CNIPA invalidation records or 中国裁判文书网. If your commercial exposure is in China, that is where an invalidity challenge would be most consequential — and it would never appear on a U.S. docket.
6. Close the residual verification gaps. The three searches I could not complete (Justia PTAB fields, IPR2023-era dockets naming State Grid Zhejiang, and U.S. reexamination/supplemental-examination records for application 16/553,665) should be run directly against PTAB E2E (https://ptacts.uspto.gov), the Patent Center reexamination record for 16/553,665, and a subscription database (Docket Navigator / Lex Machina) keyed to the patent number across all IPR/PGR/CBM/DER trial types. Treat today's finding as a negative search result corroborated by two independent sources, not as a certified absence.

Caveat: The canonical ODP block, my web checks, and the face-of-record legal-events timeline all agree on zero AIA activity. But AIA trial filings can appear in web sources before ODP ingest, and I was cut off mid-verification on three queries. If you can supply authenticated PTAB E2E access or a litigation-database export, I can convert this from "none found" to a verified negative.

I'll reconstruct the chain of title for US 11,402,438 from the assignment records and cross-reference the assignees' corporate status.

Let me query the USPTO assignment records directly via their indexed API and search for the recorded reel/frame.

The single recorded assignment (Reel 050359/0769) is visible in the legal events. Let me verify the corporate status of both assignees and identify the recording correspondent.

I hit the tool-call ceiling before I could complete direct queries against the USPTO Assignment Center API. Below is the reconstruction from what I retrieved, with the gaps explicitly flagged rather than filled.

Assignment / Ownership Reconstruction — US 11,402,438

Important provenance note: The reel/frame below comes from the Google Patents legal-events record for this patent (which mirrors the Assignment Center data), not from a direct Assignment Center query. I was unable to run the Assignment Center search UI or its API endpoint (assignment-api.uspto.gov/patent/lookup) within this session, so I could not read the correspondent-of-record field off the recorded cover sheet. Treat that one field as unverified. Everything else below is grounded in the assignment text as indexed.

Inventors

Thirteen named inventors, all individuals residing in China, per the issued patent and the recorded Assignment (Reel 050359/0769):

Employer at time of filing: Not determinable from the assignment record itself. The assignment is executed by the inventors directly in favour of the two co-assignees (see below), which is the standard pattern for an employee-invention assignment under Chinese practice where employees of a State Grid research institute and of a supplier co-file. All thirteen names appear as assignors on the same reel/frame, meaning the inventors were treated as one group transferring to two co-owners — consistent with a jointly-developed project between the utility's research arm and its equipment vendor. I cannot state which individuals worked for which entity from the four corners of the record.

Unusual-pattern check — departures: No evidence of any inventor departing, because there is no post-2019 assignment activity of any kind on this patent. There is no inventor-to-third-party or inventor-withholding assignment, no partial interest retained, and no re-recordation. No fire-sale precursor signal.

False-positive warning on inventor attribution: Third-party aggregators (e.g. patentleaderboard.com) algorithmically attach these inventors to Huawei (Jialong Xu) and Robert Bosch GmbH (Yifan He) based on name-string matches to unrelated inventors. Those attributions are not supported by this record and should not be relied on. The only employer relationship supported by evidence is the two co-assignees named on Reel 050359/0769.

Original assignee

The issued patent names two joint original assignees, both Chinese:

1. Hangzhou Kelin Electric Co., Ltd. (杭州柯林电气股份有限公司) — primary commercial owner.

• Line of business: Electric power equipment intelligent sensing; online monitoring and diagnostic early-warning digital platforms for power equipment. Its own disclosures describe product lines spanning generation/transmission/transformation/distribution and high/ultra-high/UHV voltage classes, expressly including online diagnostic and early-warning platforms for power equipment state and fault diagnosis. This is squarely the field of the patent.
• Product embodying the claims: Yes, on the face of it. The company's business description covers intelligent monitoring series products and online diagnosis/early-warning platforms for power equipment — i.e., the same subject matter as the patent's on-line winding-deformation diagnosis method. I have not been able to independently verify a specific catalog SKU implementing the claimed entropy+SVM method, so this is a strong circumstantial match, not a confirmed product-to-claim mapping.
• Current status: Operating, publicly listed. Listed on the Shanghai Stock Exchange STAR Market, ticker 688611, IPO at RMB 33.44/share, trading since 2021-04-12 (post-dating this patent's filing). Continuing operations in monitoring/diagnosis plus newer storage and perovskite PV lines.

2. Electric Power Research Institute of State Grid Zhejiang Electric Power Co., Ltd. (国网浙江省电力有限公司电力科学研究院) — research-institute co-owner.

• Line of business: Research and development arm of State Grid Zhejiang Electric Power Co., Ltd., itself a provincial subsidiary of State Grid Corporation of China (a Chinese state-owned enterprise and the world's largest utility).
• Product: Not a product-shipping entity; it is an R&D/standards/testing organisation. It tests and statistically owns such subject matter rather than commercialising it.
• Current status: Operating (SOE research institute). Not a Chapter 7/11 analogue; no bankruptcy or insolvency event relevant here.

Corporate-status contrast that matters: The two co-owners have asymmetric incentives. One is a listed equipment vendor that could plausibly assert; the other is a State Grid SOE research institute, which in practice does not assert patents offensively. That deadweight co-ownership alone substantially explains the absence of assertion activity.

Assignment timeline

Chronological list of every recorded assignment:

2019-07-17 → 2019-07-20 (executed; the record states signing dates "FROM 20190717 TO 20190720") / recorded 2019-09-12 — Reel 050359/0769

• Conveyance: Assignment — "ASSIGNMENT OF ASSIGNORS INTEREST"
• Assignor: Yiming Zheng; Wenhao Wang; Jialong Xu; and others (i.e. all thirteen named inventors, per the reel text "ASSIGNORS: ZHENG, YIMING, WANG, WENHAO, XU, JIALONG; AND OTHERS")
• Assignee: State Grid Zhejiang Electric Power Company Limited Electric Power Research Institute, China — and — Hangzhou Kelin Electric Co., Ltd., China (both named as owners on the same assignment, same reel/frame)
• Correspondent: Not captured in the indexed record I could retrieve. The correspondence field on the Assignment Center cover sheet was not available to me. For what it is worth, the attorney/agent of record on the face of the patent is Muncy, Geissler, Olds & Lowe, P.C. — but I explicitly decline to assert that they were the recording correspondent on Reel 050359/0769, because I could not read that field. This is the one item you should verify directly. Flag: Muncy Geissler is a mainstream prosecution firm, not a known NPE-side recording operation; a single appearance here would not be a signal in any event.
• Context: Initial employee-inventor assignment to the employers, executed before the US filing (2019-08-28) and roughly seven months after the CN priority filing (CN 201910091394.9, 2019-01-30). Standard clean-chain practice to support the filing, not a transfer-to-asserter, not a reorg, not a securitisation.

That is the entire chain. There are no further recorded assignments. Specifically, there is no: post-issuance assignment; security agreement; merger or change-of-name record; corrective/reel-and-frame correction; partial-interest release; or license recorded against the patent.

Do not misread the 2026-01-29 legal event. The 4th-year maintenance fee payment (M1551, large entity) appearing in the legal-events feed is a fee payment, not an assignment, and does not appear as an Assignment Center record.

Timeline diagram

timeline
    title Ownership of US 11402438
    2019 : Inventors sign assignment to both co-owners
         : US application 16 553 665 filed
         : Assignment recorded Reel 050359 0769
    2021 : Kelin Electric lists on SSE STAR Market 688611
    2022 : Patent issues as US 11402438 B2
    2026 : Fourth-year maintenance fee paid

NPE / troll-pattern signals

1. Shell-entity transfer — not present. The chain contains exactly one link, Reel 050359/0769, running from individual inventors to two identified operating entities. No "IP / Holdings / Ventures / Licensing" vehicle appears anywhere. Both assignees are named with full legal names and countries ("CHINA") on the face of the recorded assignment. No registered-agent service address, no single-member LLC of any kind.

2. Known asserter in the chain — not present. Neither current assignee — Hangzhou Kelin Electric Co., Ltd. nor Electric Power Research Institute of State Grid Zhejiang Electric Power Co., Ltd. — appears on any public NPE list (Acacia, Marathon, IV, IPNav, Wi-LAN, Conversant/Mosaid, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, DGC, Spangenberg entities, or Unified Patents / RPX high-frequency-plaintiff directories). Kelin is a publicly listed operating vendor (SSE STAR 688611), which is structurally incompatible with an NPE profile.

3. Repeat correspondent across the chain — not present, and not testable. There is only one recorded assignment, so recurrence is impossible by construction. I could not retrieve the correspondent name on Reel 050359/0769 (see gap above). The face-of-patent agent, Muncy, Geissler, Olds & Lowe, P.C., is a conventional prosecution firm; even if confirmed as recording correspondent, a single appearance is by the task's own standard not a finding.

4. Cascading transfers — not present. Zero consecutive transfers across zero chained entities within any window. One assignment in >7 years of pendency/term.

5. Pre-litigation transfer — not present. There is no infringement suit naming this patent (consistent with the litigation analysis already generated), so there is no pre-suit assignment to time against. The sole assignment pre-dates filing, which is the opposite of a pre-litigation venue/standing arrangement.

6. Bankruptcy fire-sale — not present. Neither assignee is in bankruptcy or insolvency. Kelin Electric is a going-concern listed issuer; the State Grid research institute is an SOE unit. No Chapter 7/11 analogue, no court-supervised sale.

7. Privateering — not present. No operating company has transferred this patent onward to an NPE. There is no downstream NPE at all to assert on Kelin's behalf. No SEC 10-K/8-K disclosure of a monetisation or privateering arrangement was surfaced. (Note: Kelin is an SSE-listed, not SEC-registered, issuer, so US proxy filings would not capture any such arrangement in any event — a genuine blind spot.)

8. Defensive aggregator — not present. The chain does not terminate at RPX, AST, LOT, Unified Patents or OIN; it terminates at the two original co-owners and stays there. However, the substantive effect is quasi-defensive: co-ownership by a State Grid SOE research institute alongside a vendor creates a classic "hostage patent" structure that is very rarely asserted, and never unilaterally.

Structural note (not a named signal, but relevant): joint ownership by a State Grid provincial research institute and a private vendor is a collaboration-ownership pattern common in Chinese grid-technology programmes. It is the single most predictive fact here: an SOE research institute as co-assignee will almost never join an offensive assertion campaign, and cannot be compelled to.

Verdict

Insufficient data — in the specific sense defined by the task: only the original assignment exists. Reel 050359/0769, executed 2019-07-17/20 and recorded 2019-09-12, is the sole entry in the chain, transferring all rights from the thirteen named inventors to Hangzhou Kelin Electric Co., Ltd. and the State Grid Zhejiang Electric Power Research Institute. There is no post-issuance assignment, no shell-entity transfer, and no downstream NPE, and neither assignee appears on any public asserter list.

To be clear about what the "insufficient data" label does not mean: this is not a case of an open question about whether an NPE lurks in the chain. The available evidence affirmatively points away from an NPE pattern — a listed operating vendor (SSE STAR 688611, trading since 2021-04-12) and an SOE research institute still hold the patent, they shipped (or at minimum market) the same class of monitoring/diagnosis product, and no assertion activity exists to create a pre-litigation-transfer signal. The label applies because the task's rubric reserves "operating-company assertion" for a current assignee that is suing actual competitors, and there is no suit. If you need a single sentence: not an NPE; a non-asserting, jointly-held, operating-company patent.

Verification steps still outstanding (be explicit with your client about these):

• Pull the correspondent-of-record field for Reel 050359/0769 from the Assignment Center — the one field I could not read. Link: https://assignmentcenter.uspto.gov/ (search by application number 16/553,665 or patent 11,402,438; the legacy mirror is https://assignment.uspto.gov/patent/index.html).
• Confirm whether the US assignment documents an obligation to assign pre-dating it (e.g. an employment agreement), which the recorded reel text does not disclose.
• Because any enforcement of this family would realistically occur in China, against Chinese transformer manufacturers, a US-docket-only negative is weak. Check 中国裁判文书网 (China Judgments Online) and CNIPA invalidation records against the family member CN 109884459 B, and confirm whether the two co-owners have executed any internal licence or exclusive-licence arrangement (which in Chinese practice can be recorded separately and would not appear in US Assignment Center at all).
• Kelin Electric's SSE annual reports (e.g. the 2026 interim disclosure referencing 319 IP rights and its online monitoring/diagnosis product lines) are the best available proxy for an SEC-style disclosure of any monetisation arrangement.

I'll verify the patent record at the USPTO and then pull details on every reference cited on the face of US 11,402,438.

Now let me pull details on the remaining cited references.

Now the remaining references, including the "Family Cites Families" entries and a USPTO record check.

Prior-Art Analysis — US Patent No. 11,402,438 (application 16/553,665)

1. USPTO record confirmation — exact number 11402438

I searched for the number 11402438 literally. The only record matching that exact string is:

Excluded per instruction (similar but different numbers, not this patent): US 11,402,437 / US 11,402,439-style neighbors; US 8,441,438 (CyWee v. Samsung — a different patent already flagged in my earlier section); the "US 11402437 B / US 11402438 B" strings inside EP 3082130 A1 (mangled renderings of U.S. application serial Nos. 11/402,437 and 11/402,438). None of these are US 11,402,438.

Access caveat (stated plainly): I could reach the USPTO-derived full-text PDF and USPTO-sourced mirrors (Justia, Patents Encyclopedia), but I could not authenticate into USPTO Patent Public Search / Patent Center to run a native fielded query. Everything below is grounded in the patent's own face (the front-page citation lists) as reproduced on the authoritative full text you supplied, cross-checked against the cited documents' own records. Where I could not retrieve a cited document's disclosure, I say so rather than guessing at its content.

Nothing in the citation record contradicts the two earlier sections of this analysis (bibliography, assignees, no litigation found). No contradiction to flag.

---

2. What is actually "cited" on the '438 face — and why that matters

The '438 front page carries two distinct citation lists, and conflating them is a common error:

• "Patent Citations (5)" — the five U.S. documents cited in the U.S. prosecution of 16/553,665 (each marked with an asterisk = cited by the examiner). These are the references with real §102/§103 exposure in the U.S.
• "Family Cites Families (4)" — four DE/CN documents cited in the prosecution of the family (principally the Chinese parent CN 201910091394.9). These matter for the CN counterpart's validity and as §102(a)(1) "printed publications" as of their own publication dates, but they were not necessarily before the U.S. examiner.
• ("Similar Documents" — e.g., Duan et al. 2019, Gao et al. 2022 — are algorithmic suggestions, not citations. Excluded.)

Statutory framework: The '438 has an effective filing date of 2019-01-30 (assuming the CN priority is perfected; Google expressly labels the priority date an assumption). It is a post-AIA case, so AIA §§102(a)(1)/(a)(2) and §102(b)(1)/(b)(2) exceptions govern. A reference is §102(a)(1) art if published/patented before 2019-01-30; it is §102(a)(2) art if it is a U.S. patent or U.S. application publication effectively filed before 2019-01-30, regardless of its publication date.

---

3. The five U.S. "Patent Citations" — citation, dates, description, §102 assessment

3.1 US 2020/0200813 A1 — Zhejiang University (the single most material reference)

What it discloses (from its claims and description as published):

1. Take the known-winding-state transformer's current, voltage, current differences and voltage differences per phase of each winding as on-line monitoring indicators, and group them into position sub-samples;
2. Split the on-line monitoring data into two sequences by time and normalize them to dimensionless sequences;
3. Compute permutation entropy, wavelet energy and arithmetic mean of each of the two sequences, and compute root-mean-square errors of each;
4. Build a four-dimensional feature set from the three RMS errors plus a cumulative short-circuit current for the corresponding position sub-sample;
5. Add a label (deformed / not deformed) and input into an SVM to train a diagnostic model;
6. Extract the same feature set from a transformer under diagnosis, input to the trained model, and diagnose each position sub-sample for winding deformation.

It expressly enumerates the 9 position sub-samples as "high-voltage phase-A, high-voltage phase-B, high-voltage phase-C, medium-voltage phase-A, medium-voltage phase-B, medium-voltage phase-C, low-voltage phase-A, low-voltage phase-B, and low-voltage phase-C" — verbatim the same nine positions as '438 claim 2.

§102 exposure map against '438 claim 1:

Verdict — potential anticipation: US 2020/0200813 A1 is the closest prior art and the only cited reference with realistic §102(a)(2) anticipation exposure. On the record I retrieved, it does not appear to disclose every limitation of claim 1 (Step 4 and the Step 5 gating), so I would characterize it as (i) a strong §103 reference (its differences are, on their face, routine implementation choices), and (ii) potentially anticipatory of a narrower claim if claim 1's Step 4/Step 5 language were read as inherent — which I cannot confirm without its full disclosure.

Critical legal point (please don't skip): an independent claim's limitations carry into every dependent claim. If '813 does not anticipate claim 1, it cannot anticipate claim 2 (or 3–7) even though claim 2's nine positions appear verbatim in '813 — a dependent claim is anticipated only if the reference discloses all limitations of claim 1 plus the added limitation. Anyone asserting "claim 2 is anticipated by '813" would be making that error.

Ownership/inventorship nuance worth flagging: '813's assignee is Zhejiang University; the '438's are Hangzhou Kelin Electric and State Grid Zhejiang EPRI. However, Hua, Zhongsheng (华中生) appears among the inventors of both the '438 and the CN109444656 family, and is listed in the inventors' IET RPG-2019 paper as affiliated with Zhejiang University. That means the §102(b)(2)(C) common-ownership exception is not available (different owners, no common obligation of assignment on this record), but there is a genuine inventorship-overlap / derivation-of-subject-matter question that is analytically separate from §102 and that I flag for follow-up. I have not verified that the two "Hua, Zhongsheng" entries are the same person.

3.2 US 2013/0132001 A1 — Yacout (Polyvalor)

Description: A condition-based-maintenance tool: a database of measured indicators of a device's dynamic condition; a binarization module; and a machine-learning data-mining module using Logical Analysis of Data (LAD) with Mixed Integer Linear Programming pattern generation to extract patterns indicative of fault/no-fault, then a discriminant function used for diagnosis and prognosis. It supports multi-classification and unsupervised learning. Its worked examples are power-transformer-related (DGA fault typing — thermal vs. energy-discharge defects, Rogers-ratio rules) and bearings; FIG. 12 shows a Daubechies wavelet transform of a bearing signal; its background expressly surveys SVM and ANN/FFNN as known fault-diagnosis techniques, and compares multilayer LAD against SVM classifiers.

§102(1) assessment: Not anticipatory of any claim of the '438. It discloses no permutation entropy, no wavelet entropy as a transformer-winding feature, no three-phase three-winding 9-position subsample decomposition, no pre-/post-short-circuit RMSE feature set, and no winding-deformation positioning. Its transformer examples concern dissolved-gas analysis, not winding geometry.

Role: §103 background art for (a) "machine-learning classification is a known way to detect device faults, including in power transformers," and (b) "SVM is a known classifier in device fault diagnosis." It is the reference an examiner would pair to show the ML/SVM architecture was conventional — which supports the reading that '438's novelty resides in the entropy feature set + 9-position decomposition, not in "use an SVM."

3.3 US 2014/0039817 A1 — Phase3 Technologies Ltd. (Prosper Dayan)

Description: On-line monitoring of an electrically-connected system with periodic behavior (motors, generators, pumps, turbines) using two-phase current sensors (Rogowski coils). It transforms the measured currents into "initial current information" — a normalized correlation frequency spectrum, a non-trivial frequency spectrum, and a root-mean-square calculation of that spectrum — performs an initial threshold check of normal operation, then trains/models the system to yield "modeled current information" (coefficient vector + training feature table), and compares "instant current information" to the model by computing a residual energy against thresholds to determine operating status. Fault classes are mechanical (shaft imbalance, bearing spalling, blade fracture, housing deformation, etc.).

§102(1) assessment: Not anticipatory of any claim of the '438. No transformer winding deformation, no permutation entropy, no wavelet entropy, no 9-position decomposition, no SVM, no pre/post-short-circuit sequence comparison. Notably, the "housing fracture/deformation" and RMS-of-spectrum language are superficially close to the '438 vocabulary but are directed to rotating machinery, not three-phase power-transformer windings.

Role: §103 background art for (a) on-line (in-service) monitoring of power apparatus from current signals without outage, and (b) the general "compare a current-state feature set to a trained/modeled baseline and threshold the residual" framework. This is the reference supporting a motivation-to-combine argument that "doing this on-line rather than off-line was known."

3.4 US 2020/0103894 A1 — Strong Force IoT Portfolio 2016, LLC

Description: A very large industrial-IoT portfolio family. It generates streams of industrial-machine health-monitoring data by applying machine learning to data representative of machine portions received over a data-collection network, produces service recommendations via industrial machine fault detection and classification algorithms, and predicts service events from vibration data captured by vibration sensors, generating a maintenance-action signal based on a computed severity unit; a CMMS (computerized maintenance management system) consumes the output. Classification art cited includes PCA/PLS, Bayesian networks, linear regression/correlation, and neural networks (see its CPC set: G05B23/024, G06N3/02, etc.).

§102(2) assessment: Not anticipatory of any claim of the '438. It contains no permutation entropy, no wavelet entropy, no 9-position three-phase three-winding decomposition, no pre/post-short-circuit RMSE feature set for transformer windings, and no SVM classification of winding-deformation position. Its "machine signals"/"vibration data" are general industrial machinery, not transformer-winding deformation.

Role: §103 background art (at most) for "it was known to apply machine learning to machine-monitoring signals for computerized maintenance management." It is the kind of reference an examiner cites to show the generic ML-on-machine-signals environment, not the '438's specific technical contribution.

3.5 US 2021/0048487 A1 — Wuhan University

Description (as titled): transformer winding fault positioning using a deep convolutional neural network integrated with visual identification. Doctrinally it is the "same problem, different tool" reference (deep CNN + vision vs. permutation/wavelet entropy + SVM).

§102 assessment: Because it post-dates the '438 effective filing date, it cannot anticipate any claim. I list it here only because it appears on the face of the patent and a careless analyst might treat the listed citation as automatically-available prior art. It is not. (It would only become relevant if the '438's CN priority were held unperfected and the effective date moved to 2019-08-28 — still after 2019-08-12, so still not art. It stays out of the §102 analysis either way.)

---

4. The four "Family Cites Families" references

These were cited in the family's (Chinese) prosecution, so they are §102(a)(1) "printed publications" by their own publication dates and are available as art against the '438 as a matter of date — but they were not necessarily before the U.S. examiner, and I could not retrieve their full disclosures, so my §102 assessment is deliberately partial.

---

5. Ranking: most relevant prior art for US 11,402,438

1. US 2020/0200813 A1 (Zhejiang University) — by a wide margin. §102(a)(2) art; discloses the 9-position three-phase three-winding decomposition, permutation entropy, wavelet (energy/entropy), arithmetic mean, RMSE feature construction, SVM training with deformation labels, and per-position diagnosis. Strongest §102(a)(2) candidate; strongest §103 reference. Its apparent gaps vs. claim 1 are Step 4 ("hierarchical cross-validation" yielding accuracy/precision/recall) and the Step 5 detect-then-localize gating, plus the "before and after the last short circuit" vs. "according to time" sequence-splitting wording, and "wavelet entropy" vs. "wavelet energy."
2. CN 105627904 B (State Grid Zhejiang EPRI). §102(a)(1); same assignee, same technical subject (winding-deformation determination). Relevance to the Step 5 diagnosis branch is high; disclosure unverified.
3. CN 103163420 B (Shenyang Univ. of Technology). §102(a)(1); "intelligent on-line transformer state judgment." §103 context; disclosure unverified.
4. US 2013/0132001 A1 (Yacout/Polyvalor). §102(a)(1); ML (LAD) fault detection with transformer DGA examples and an express SVM/ANN background survey. §103 architecture reference only.
5. US 2014/0039817 A1 (Phase3). §102(a)(1); on-line current-based monitoring with trained baseline + RMS/residual comparison. §103 "on-line was known" reference.
6. US 2020/0103894 A1 (Strong Force IoT). §102(a)(2); generic ML-on-machine-signals for maintenance. §103 environment reference.
7. CN 105956623 A (fuzzy-entropy EEG classification). §102(a)(1); §103 background for entropy + classification.
8. US 2021/0048487 A1 (Wuhan University). Not prior art (post-dates the '438 effective filing date). Excluded from §102.
9. DE 4445794 C1. Unrelated art; no claim element disclosed.

No single cited reference, on the record I could retrieve, discloses every limitation of '438 claim 1. Accordingly, on this record I would not assert §102 anticipation of claim 1 — or, by derivation, of claims 2–7 (each of which incorporates claim 1's limitations). The realistic attack on the '438 is §103 over US 2020/0200813 A1 alone or in view of Yacout / Phase3 (for the SVM architecture and on-line monitoring), with CN 105627904 B supplying the Step 5 diagnosis branch.

---

6. A defensive point worth recording: the inventors' own publications do not appear to be prior art

Two same-team publications surface in this field:

• IET RPG 2019, "On-Line Diagnosis Method for Transformer Winding Deformation Based on Running Voltage and Current Correlation Mining" (Zheng, Wang, Gong, Li, Sun, Du, Hua) — https://digital-library.theiet.org/doi/abs/10.1049/cp.2019.0328
• IEEE EI2 2019 (Changsha, 8–10 Nov 2019), "Online Diagnosis for Transformer Winding Deformation Based on Information Entropy and SVM" — DOI 10.1109/EI247390.2019.9062191

Both describe the same permutation-entropy + wavelet-entropy + SVM method, and both post-date the '438's 2019-01-30 effective filing date (the EI2 conference was 8–10 Nov 2019; IET RPG 2019 was also held in late 2019). On that basis they are not §102(a)(1) art against the '438. I flag one uncertainty: I have not independently confirmed the IET RPG 2019 publication (as opposed to presentation) date, so if the '438's priority date were ever challenged, this should be re-verified. These papers are, however, highly relevant as corroboration of the '438's disclosure and as a §103 obviousness/§102 mapping aid in any later dispute.

---

7. Gaps, uncertainties and recommended verification

• US 2020/0200813 A1 full disclosure not fully retrieved. I have its published claims and partial description. To convert my "appears not to disclose Step 4/Step 5" into a firm statement I need its complete description and the CN 109444656 spec (小波熵 vs. 小波能量 wording matters for literal §102).
• CN 105627904 B and CN 103163420 B full texts not retrieved. Both are §102(a)(1) art by date; F3 (CN 105627904 B) is the one most likely to matter, given its identical subject matter and shared assignee.
• Priority-date assumption. Google expressly labels the 2019-01-30 priority date an assumption. If CN 201910091394.9 does not support the '438 claims, the effective date shifts to 2019-08-28 — which would not change the §102(a)(2) availability of US 2020/0200813 A1 (effective 2018-12-21) or US 2020/0103894 A1 (effective 2018-05-07), but would change the availability analysis for anything published between those dates.
• §102(b)(2)(C) not available on this record (different owners), notwithstanding the apparent inventor overlap (Hua, Zhongsheng / 华中生) with the Zhejiang University family — a separate derivation/§115 issue, not a §102 exemption.
• "SVM" expansion. Consistent with my earlier section: the patent text renders SVM once as "Small Vector Machine"; that is a translation artifact. It does not change the §102 analysis, since the cited art (Yacout's background; '813's claims) uses "support vector machine" in the ordinary sense.

Bottom line: Searching the number 11402438 yields exactly one record — US 11,402,438 B2. Of its nine front-page citations, US 2020/0200813 A1 (Zhejiang University, §102(a)(2)) is the most relevant and the only one with genuine anticipation potential, but on the retrieved record it does not fully meet claim 1; US 2021/0048487 A1 is not prior art at all; and the remaining references are, at most, §103 material.

I'll analyze the obviousness of US 11,402,438 using the prior art of record cited on the patent face. Let me first pull details on the key references.

Now let me verify a critical prior-art-status fact: the inventor overlap between US 2020/0200813 A1 (Zhejiang University) and the '438 patent, which affects the § 102(a)(2)/§ 102(b)(2) analysis.

Obviousness Analysis — U.S. Patent No. 11,402,438 Under 35 U.S.C. § 103

Scope note / cross-reference: This section builds on the earlier-generated sections (claim structure, no-litigation finding, PTAB-virgin status) and does not repeat them. Where the earlier sections assumed the '438 patent's prior-art citations of record, I test that assumption below and flag one discrepancy. All claim references are to the claims as issued (7 claims: claim 1 independent; claims 2–7 dependent). Date discrepancy worth noting: the task block states April 26, 2026; the system date is 2026-09-30. Nothing in this analysis turns on it.

---

1. Bottom line up front

The '438 patent is, in my assessment, highly vulnerable under § 103 — and arguably under § 102 — because a near-identical, earlier-effectively-filed U.S. pre-grant publication is already on the face of the patent: US 2020/0200813 A1 (Zhejiang University, "Online diagnosis method for deformation position on trasnformation winding"), which is § 102(a)(2) art as to the '438 patent's 2019-01-30 priority date. Its U.S. claims recite the same six-step architecture (9 position subsamples → permutation entropy/wavelet feature + arithmetic mean → RMSEs → label → SVM → per-subsample output). The '438 patent differs principally by (a) reciting "wavelet entropy" where the '813's English translation says "wavelet energy," (b) adding a pre-gate step ("first carrying out an on-line winding deformation diagnosis method to conclude that the transformer to be tested is deformed"), and (c) adding model-validation metrics ("hierarchical cross-validation… accuracy, precision, and recall") and 30-indicator bookkeeping. Each of those three differences is bridged by the art of record plus the patent's own admissions.

Two important structural facts sharpen this:

• The '813 was cited by the examiner on the face of the '438 patent (the "Citations (5)" block), yet the claims issued after only one non-final Office action (mailed 2022-02-09), one response (2022-03-05), and a notice of allowance (2022-05-13) — the chronology from the Legal Events timeline at https://patents.google.com/patent/US11402438/en. Whatever distinction the applicant argued in that single response is the crux of the merits case and must be pulled from the file wrapper.
• There is a probable inventor overlap. The Chinese counterpart of the '813 is CN 109444656 B (浙江大学 / Zhejiang University, filed 2018-12-21, published 2019-03-08, granted 2020-06-09), whose listed inventors are 华中生、游雨暄、徐晓燕 (see the first page of https://patentimages.storage.googleapis.com/27/3b/fe/81de4909dbc326/CN109444656B.pdf). "华中生" strongly suggests Zhongsheng Hua, who is a named inventor on the '438 patent. I flag this as a hypothesis, not a verified fact — it materially affects the § 102(b)(2) analysis in § 9 below.

---

2. Governing law and level of ordinary skill

• AIA applies. The '438 patent claims priority to CN 201910091394.9 (2019-01-30) and was filed 2019-08-28 — both after 2013-03-16. §§ 102/103 as amended govern.
• Effective filing date (assuming perfected priority): 2019-01-30. This is the date against which art must pre-date.
• POSITA: a degreed electrical power engineer (B.S./M.S.) with ~2–5 years in power-transformer condition monitoring, or a signal-processing/ML engineer working on electrical asset diagnostics, with working knowledge of (i) transformer on-line current/voltage monitoring, (ii) permutation entropy and wavelet-based complexity measures, and (iii) supervised classification (SVM) and its validation. This is a combination skill set; the '438 patent's specification itself jumps freely between the two, which is itself evidence that combining them was routine.
• Framework: KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007) (predictable variation; known technique; finite number of identified, predictable solutions); MPEP 2143/2144 (rationale for combining, "obvious to try," design incentives). No teaching, suggestion, or motivation in the references is required.

---

3. Prior-art status of the references (dates and § 102 basis)

This is where most § 103 analyses of this patent go wrong, so I set it out explicitly. The '438 patent's face lists five U.S. publications as "cited by examiner" and four foreign family citations.

Key takeaway: the only reference that discloses the claimed combination itself is the '813, and the '813 is only 102(a)(2) art. That makes the § 102(b)(2) exceptions — not obviousness doctrine — the decisive battleground for the strongest ground.

---

4. Ground 1 (headline): Claim 1 obvious over US 2020/0200813 A1 (primary), optionally in view of Yacout '001

Element-by-element. Quotations are from the '813 as published (https://patents.google.com/patent/US20200200813A1/en; https://patents.justia.com/patent/20200200813).

What the '813 additionally has that the '438 does not claim — a fourth feature, "cumulative short-circuit current." That is a broadening by the '438, which cuts against the patentee (In re NTP; deleting a disclosed element is a classic obviousness posture).

Motivation (Ground 1): the '813 is the same invention in the same field, solving the same problem by the same architecture. Standing alone that is sufficient under MPEP 2144.09 (same field of endeavor / same problem); where the primary reference differs only in unclaimed detail and in two conventional implementation choices, the combination rationale is "applying a known technique to a known method, yielding predictable results" (KSR).

---

5. Grounds 2–4: alternative and stacked combinations

Ground 2 — '813 + Yacout '001. Yacout teaches exactly the classifier-training and evaluation layer that is the '438's Step 3–4 contribution:

• "store in a database a plurality of measured indicators representative of at least one dynamic condition of the device"; "binarize"; "analyze… using a machine learning data tool for extracting at least one pattern… indicative of whether the device has a fault or not";
• multi-classification; discriminant function; and a dedicated flow-chart block "EVALUATION OF THE ACCURACY OF CLASSIFICATION" with "TEST DATA" (see https://patents.google.com/patent/US20130132001A1/en);
• power-transformer worked examples (DGA: energy-discharge vs. thermal defects; Rogers ratio rules), i.e., the same asset class;
• a wavelet feature-extraction step ("Daubechies Wavelet Transform (DWT) of signal…") and an express SVM comparison ("training time for Multilayer LAD and SVM based classifiers…").
  Motivation: the reference identifies a finite set of known classifiers (LAD vs. SVM) for the same task and expressly benchmarks SVM — the definition of "obvious to try" with predictable results (KSR; MPEP 2144.07).

Ground 3 — '813 + Phase3 '817 (US 9,316,676 B2). Phase3 teaches on-line, sensor-based monitoring of an electrically-connected, polyphase, periodic system: two current sensors on two phases; transforming/normalizing measured currents; root-mean-square computation; residual-energy comparison against a threshold; and a train-then-model-then-monitor sequence (https://patents.justia.com/patent/[9316676](/patent/9316676)). Motivation: supplies the "on-line monitoring indicators" acquisition and RMS-feature machinery and the "normal-baseline-then-monitor" metrology for a two-/multi-phase electrical asset — a known technique used to improve a similar device in the same way.

Ground 4 — '813 + Strong Force '894. Strong Force teaches an "industrial machine predictive maintenance facility that produces industrial machine service recommendations… by applying machine fault detection and classification algorithms," machine learning applied to machine-sensor data, and expressly enumerates "support vector machines (SVMs), Bayesian network, … deep learning, … KNN" among the ML options, with industry-specific feedback training (https://www.freepatentsonline.com/y2020/0133255.html). Motivation: confirms that, as of the priority date, applying a standard supervised classifier to industrial-machine signals for fault detection/classification was a routine, off-the-shelf engineering choice — i.e., the '438's Step 3 is a "familiar element."

Ground 5 — the entropy-algorithm claims as textbook. Claims 4 (permutation entropy) and 5 (wavelet entropy) recite named, published algorithms:

• permutation entropy: phase-space delay-coordinate reconstruction, ordinal pattern counting, relative frequency P(i), H = −ΣP·ln P, normalized by ln(m!) — Bandt & Pompe, "Permutation entropy: a natural complexity measure for time series," Phys. Rev. Lett. 88, 174102 (2002) (citation from my knowledge; verify);
• wavelet energy entropy: multiresolution decomposition into D₁…D_M and A_M(n), band energies, relative energies e_i, H = −Σe_i·ln e_i — Rosso et al., "Wavelet entropy…," J. Neurosci. Methods 105, 65–75 (2001) (verify);
• the '813's CN counterpart recites "小波熵" (wavelet entropy) in claim 1 step (3) — see the CN 109444656 B claims at the PDF link above — which directly rebuts any argument that the '813 discloses only "wavelet energy."

Motivation: where a dependent claim recites a known algorithm verbatim from the literature, the claim adds nothing patentable over a reference that uses the named technique or its near-equivalent; a "known technique used to improve a similar device in the same way" (KSR).

Ground 6 — the family-citation route (fallback if the '813 is disqualified). If the '813 falls away (see § 9), claim 1 can be assembled from: CN 103163420 B (intelligent on-line judgment of a power transformer's state) + CN 105627904 B (determination of transformer winding deformation) + CN 107037314 A (transformer winding state from wavelet-packet energy entropy compared to thresholds) + Yacout '001 (tagged ML fault classification with accuracy evaluation) + the Bandt-Pompe / Rosso algorithms + the '438's own admissions. Motivation: each reference is in the same field or a directly analogous signal-diagnosis field; CN 107037314 A supplies the express entropy-for-winding-deformation teaching; Yacout supplies the classifier. The combination is a predictable aggregation of known elements.

---

6. Dependent claims 2–7

Practical consequence for a challenger: because claim 7 defines the "on-line winding deformation diagnosis method" of claim 1's Step 5 as steps (a)–(f), and each of (a)–(d) and (f) tracks the '813, the "pre-gate" difference in claim 1 largely dissolves — claim 1's Step 5 is the '813's method run at the transformer level before running it at the subsample level.

---

7. Why a POSITA would have combined these (articulated rationale)

1. Same field, same problem, same architecture. Every reference in Grounds 1–4 addresses either transformer condition assessment or machine-fault classification from monitored electrical/vibration signals. MPEP 2144.09.
2. Express suggestion in the primary reference. The '813's background states that "another feasible way of online diagnosis method is to directly analyze the measurable signal by using signal processing techniques such as wavelet transform and Fourier transform, extract characteristic values such as amplitude, variance and information entropy, and then combine with the classifier to detect the fault type." That is a written invitation to do exactly what claim 1 recites.
3. Art-recognized entropy-for-winding-deformation teaching. The '813 discusses CN 107037314 A, which computes energy entropy of transformer winding vibration signals and compares to thresholds to determine winding state. An artisan reading the primary reference is therefore led to try additional entropy measures (permutation, wavelet) on winding-deformation signatures.
4. Art-recognized classifier-selection teaching. Yacout not only uses ML pattern extraction to answer "has a fault or not" but compares LAD against SVM for the same machine-diagnosis problems — a finite set of identified solutions with predictable results (KSR).
5. Known metrology for on-line polyphase monitoring. Phase3 shows the conventional on-line acquisition and RMS/residual-baseline technique for polyphase electrical assets; the '438's "on-line monitoring indicators" are exactly those quantities.
6. Applicant's own admissions. The '438 specification: (i) the only named prior approaches are the three off-line tests (FRA, LV short-circuit impedance, dielectric-loss capacitance); (ii) the technical problem is framed purely as an efficiency/availability problem ("without interference with the normal operation of the power grid… saving manpower and material resources"); (iii) claims 4–6 recite textbook entropy/RMSE formulas without any asserted discovery in the mathematics. Where the invention is a known technique applied to a known problem with predictable results, § 103 is satisfied (KSR).
7. Reasonable expectation of success, demonstrated. The '813's own example (27 subsamples from three transformers) and the '438's FIG. 4 scatter (deformed subsamples clustering upper-right, normal lower-left) show the entropy-difference separability. Both documents report that it worked — i.e., the results were predictable, not unexpected.
8. Design incentive / market force. Avoidance of outages, the one-year test interval problem, and labor cost — all recited in the '438 background — are classic design incentives under KSR.

---

8. Weak points, gaps, and the strongest patent-owner rebuttals

A defensible § 103 position must concede and address these:

1. "Wavelet energy" vs. "wavelet entropy." The English-language '813 describes "wavelet energy" while the '438 claims "wavelet entropy" (claim 5) and "RMSE of … wavelet entropy" (claim 1). Rebuttal: the Chinese counterpart's claim text uses 小波熵 (wavelet entropy); the '438's own formula is merely the Shannon entropy of the relative band energies; and the Rosso literature makes the step conventional. Action: pull the '813's actual U.S. specification paragraphs (not just the claims/abstract) to see whether "entropy" appears; if the U.S. spec says only "energy," rely on § 103, not § 102.
2. "Two sequences before and after the last short circuit" vs. "two sequences according to time." Rebuttal: the '438's own claim 7(b) recites the alternative rule (first half/second half when no short circuit occurred), which shows the division is an administrative choice; the '813's embodiment and the '438's application example use the same datasets/timeline (HB: 2013-11-01 → 2015-01-24 → 2015-08-13), indicating a common disclosure origin. Expect this to be a § 103, not § 102, point.
3. "Hierarchical cross-validation… accuracy, precision, and recall." The clearest textual gap; the '813 as published does not recite validation metrics. Rebuttal: Yacout's test-set/accuracy-evaluation flow plus the notorious conventionality of cross-validation (Kohavi, "A study of cross-validation and bootstrap for accuracy estimation and model selection," IJCAI 1995 — verify) and of precision/recall as classification metrics. The '438's own definition ("sequentially selecting one set as a test set and the other two sets as training sets") is ordinary 3-fold cross-validation, not a special "hierarchical" invention — which also creates a § 112 indefiniteness angle if "hierarchical" is asserted to mean something more.
4. The Step 5 pre-gate. As noted, claim 7 largely converts this into a restatement of the same method; and the '813's background explicitly contemplates the first-stage determination ("determining winding deformation by the correlation between current and capacitance"; signal-based fault-type detection) as a known capability. A § 103 rationale is "performing a preliminary determination to identify candidate transformers before running the localization model is a routine workflow optimization," and a § 112(b) rationale exists if the pre-gate's "on-line winding deformation diagnosis method" is indefinite.
5. Strongest available "teaching away" candidate (weak). The November 2019 EI2 paper notes that "it is difficult to link information entropy with the cause of the fault" and that "the lack of theoretical support makes the diagnosis process not able to be explained scientifically, which limits its application in practice." This fails as teaching away: it post-dates the priority date (so it is not art at all — the patent owner cannot rely on it as art, and a challenger should not plead it), and mere absence of a theoretical explanation does not teach away from an operable technique — the same paper proceeds to implement it, as does the '813 and the '438.
6. Secondary considerations. None is presently evidenced. If a patent owner asserts commercial success, nexus must be to the claimed combination — difficult where the claim's novelty is a known entropy-plus-SVM pipeline.

---

9. Critical caveat: the '813 may be disqualified — the § 102(b)(2) problem

Because US 2020/0200813 A1 is only § 102(a)(2) art (pre-priority effective filing, post-priority publication), it is subject to three exceptions:

• § 102(b)(2)(C) (common ownership): Fails — the '813's subject matter was owned by Zhejiang University, whereas the '438 was owned by Hangzhou Kelin Electric Co., Ltd. and State Grid Zhejiang EPRI. Not the same person; not under a common obligation of assignment.
• § 102(b)(2)(A)/(B) (obtained from the inventor / inventor's own prior public disclosure): This is the live risk. The '438's inventor list includes Zhongsheng HUA, and the CN counterpart CN 109444656 B lists 华中生 among its three inventors. If 华中生 = Zhongsheng Hua (I have not verified this identity; name-matching across transliterations is unreliable), the patent owner can argue the '813's subject matter was obtained directly or indirectly from a joint inventor of the '438 application, invoking § 102(b)(2)(A) to remove the '813 as prior art. A challenger's counter is that the '813's specific "four-dimensional feature set + cumulative short-circuit current" contribution may originate from the other two named inventors (游雨暄 / 徐晓燕), so at least that subject matter remains § 102(a)(2) art and remains combinable.
• § 102(b)(2)(B): apposite only if the '438 inventors themselves publicly disclosed the subject matter before the '813 was effectively filed — on these dates, unlikely.

Consequence: the strongest ground may rise or fall on an inventorship/derivation fact question. Before investing, verify (i) the U.S. '813's inventor list and its U.S. filing date and priority chain, (ii) whether 华中生 and "Zhongsheng HUA" are the same person, and (iii) whether Zhejiang University's application and the '438 application share a common upstream disclosure. Also note the adjacent (non-§ 103) risks this creates: derivation / correction of inventorship (35 U.S.C. §§ 115, 135, 256) and a possible ownership dispute — matters that could matter more than any invalidity theory.

Fallback if the '813 is disqualified: attack the priority claim. If the '438's U.S. claims 1–7 are not fully supported by CN 201910091394.9 (e.g., undefined "average," the "hierarchical cross-validation" limitation, or the two-stage gating), the effective filing date slides to 2019-08-28. That would (a) make CN 109444656 A (published 2019-03-08) available as § 102(a)(1) art, and (b) make US 2021/0048487 A1 (Wuhan University, effectively filed 2019-08-12) available as § 102(a)(2) art. Neither route helps if priority holds, so this is a contingency, not the primary theory.

Caveat on the '813 as a § 102 reference. I charted claim 1 against the '813's published claims, summary, and abstract, which I retrieved in full, plus the CN counterpart's claim text. I have not read the '813's complete U.S. specification (paragraph-level disclosure, figures, examples). A complete § 102/§ 103 chart must be built from the specification itself.

---

10. Discrepancies to flag against the earlier sections

1. Prior-art composition. The earlier PTAB section stated the five cited pre-grant publications were "all cited in the original IDS." I confirm five U.S. publications are on the face of the patent, but US 2021/0048487 A1 is not prior art against the 2019-01-30 priority date (effectively filed 2019-08-12; published 2021-02-18). Either the examiner cited it as background, or the effective filing date was misapprehended. Do not build a § 103 theory on the Wuhan reference.
2. Citation noise. DE 4445794 C1 ("Gas-duct inner tube externally joined to heavy metal formers") is facially irrelevant to transformer diagnostics and should be treated as unusable, whatever its label in the family-citation table.
3. Consistency elsewhere. The earlier sections' findings (no litigation; zero AIA proceedings; claims 1–7 all untested) are unaffected. Indeed, the "PTAB-virgin, never asserted, single-Office-action prosecution" profile is consistent with a patent whose closest art was already of record: an examiner-cited near-identical reference plus a single allowance suggests the applicant narrowed the claim to something the examiner accepted as distinct — which is precisely the argument the file wrapper will contain.
4. "Small Vector Machine." The '438 specification's rendering of "SVM" as "Small Vector Machine" is a translation artifact (already flagged in the earlier summary); it is not prior art and has no § 103 significance, though it is a § 112 clarity irritant.

---

11. Recommended next steps

1. Obtain the file wrapper for 16/553,665 — specifically the 2022-02-09 non-final Office action and the 2022-03-05 response. Any amendment or argument that distinguished the applicant's disclosure from US 2020/0200813 A1 defines the ground you must meet; it may also contain a prosecution-history estoppel or § 112 hook.
2. Retrieve the complete U.S. specification of US 2020/0200813 A1 and confirm (a) the treatment of "wavelet energy/entropy," (b) whether the before/after-last-short-circuit division is described, and (c) whether any model-validation (cross-validation, precision/recall) is described.
3. Resolve the inventorship overlap. Compare the U.S. '813's inventor list against the '438's 13 inventors; determine whether Zhongsheng Hua = 华中生; and obtain the assignments/employment-obligation records of the Zhejiang University inventors. This is the pivot for § 102(b)(2)(A).
4. Confirm publication dates for CN 107037314 A and CN 103163420 B, and obtain their relevant passages if they are to be pleaded.
5. Assemble the classical-art exhibits (Bandt & Pompe 2002; Rosso et al. 2001; a cross-validation/metrics reference such as Kohavi 1995) with verified bibliographic details, since claims 4–6 will be attacked as conventional algorithm recitations.
6. Build the chart on the '813 as primary, with Grounds 2–4 pleaded in the alternative, and be prepared for the patentee to contest only the three discrete differences identified in § 8 — "wavelet entropy," the short-circuit-based sequence division, and "hierarchical cross-validation" — none of which survives a well-supported § 103 rationale.

Final caveat: this analysis is a § 103 opinion built from the patent's own text, the cited references' published front matter/summaries/claims, and targeted searches; it is not a validity opinion, and I have not reviewed the complete specification of every reference, the file wrapper, or any Chinese-language validity record. Treat the "obvious" conclusions as strong but needing paragraph-level verification.