专利申请审查反馈

审查意见通知（非正式）

一、总体结论 基于申请人提交的题目、背景技术、发明内容与权利要求的现有表述，本案属于工业过程控制领域中的制造方法，原则上不属于排除客体，具备专利客体适格性（中国专利法第2条）。然而，从现有公开内容看，权利要求存在以下主要审查风险：

• 权利要求清楚性与支持性不足（专利法第26条第4款，审查指南第二部分第二章3.1/3.2）：核心技术特征用语模糊，关键技术手段与参数缺乏界定，控制目标“熔深d”的在线测量或估算机理未披露，导致权利要求的技术效果无法核实。
• 说明书充分公开不足（专利法第26条第3款）：未充分公开1D CNN的训练与部署方案、信号预处理、实时性实现路径、控制律等，使本领域技术人员难以在不付出创造性劳动的情况下实现“≤2 ms闭环延迟”且“将熔深维持在阈值以内”的结果。
• 新颖性/创造性存在显著挑战（专利法第22条）：激光焊接过程的同轴光学与声发射传感采集、基于时序信号的深度学习质量判别、以及基于判别结果的焊接功率/能量闭环调节，均为本领域近年普遍公开与工程化趋势。申请文本未能清楚表明区别于现有技术的技术特征组合及其可验证的技术效果。

二、针对权利要求的具体意见

1. 权利要求1（方法）

• 关于C1“在焊接头同轴集成光电与声发射传感器，采集I(t)、S(t)”：
  • “同轴集成”的具体实现方式不明（透射/反射光路、分光元件、带宽与视场、抗干扰结构）。I(t)、S(t)的物理量含义不清（等离子体辐射强度？特定波段光强？声发射频带与灵敏度？）。
  • 建议限定：光学信号的波段或光学链路构成、声发射传感器类型与安装位置、采样/前端放大与抗混叠滤波参数。
• 关于C2“以轻量一维CNN…滑动窗口…熔池状态分类”：
  • “轻量”“滑动窗口”“熔池状态”均属相对性用语，边界不清。需界定：窗口时长、步长、输入通道、网络层数/核大小/参数量上限，分类标签定义（如稳定贯穿/欠熔/喷溅/孔洞先兆等），以及推理阈值。
  • 建议在从属权利要求中限定至少一组可实施的网络结构与窗口配置，以满足清楚性与支持性。
• 关于C3“自适应调节脉冲能量E与占空比D，使熔深d保持在[dmin,dmax]”：
  • 申请文件未交代在线“熔深d”的测量或估算路径。若仅凭I(t)、S(t)分类，无法直接得到d，应说明由分类结果/回归模型映射至d的算法与标定流程。
  • “自适应调节”的控制律未披露（如PID/MPC/增益调度/查表），稳定性与响应时间如何保证以满足“控制延迟≤2 ms”。建议在权利要求中至少限定控制更新周期与关键增益或决策逻辑框架。
• 关于C4“预测‘孔洞风险’时执行跳点策略并记录二维码追溯数据（包含轨迹哈希与特征向量）”：
  • “孔洞风险”的判定阈值或指标未定义；“跳点策略”含义不明（跳过当前点、重新排序路径、或延时重复焊接），与焊接节拍、热输入的关联缺失。
  • “二维码追溯数据（轨迹哈希与特征向量）”属于数据组织/展示表述，若无与过程控制、设备自检或后续闭环的因果关联，易被认定为非技术性特征。建议将追溯数据与后续工序的自动参数化或合格判定联动，体现可验证的技术效果。
• “采样频率≥200 kHz、控制延迟≤2 ms”：
  • 作为限定特征有助于区分部分现有技术，但需说明实现路径（如FPGA/ASIC或低时延DSP、DMA零拷贝、固定点推理等）。否则存在第26条第3款之风险。

2. 权利要求2（追溯数据与电芯序列号绑定）

• 该特征偏向数据管理/业务规则，缺乏直接技术效果的限定。建议将“绑定”与具体的生产过程控制或质量判定阈值自适应关联起来，以增强技术性与创造性贡献。

三、说明书充分公开与支持性之不足 根据专利法第26条第3款与审查指南第二部分第二章3.1：

• 未公开训练数据的来源、标注口径（熔深/孔洞的金相或CT验证基准）、域自适应方法（不同极耳材质与厚度）、防过拟合策略以及边界条件，从而难以复现“适配不同材料与厚度”的普遍性效果。
• 未公开信号链路（前置放大、滤波、同步、去噪、归一化）、特征工程（若有）与推理加速方法，无法保证“≤2 ms”的端到端时延。
• 未公开“熔深d”的在线估算机理与标定流程（例如以试板建立I(t)/S(t)到d的映射、在线漂移补偿），导致C3之效果缺乏可操作性支撑。
• “跳点策略”未给出确定性流程（触发条件、最小跳转间隔、恢复策略、对热累积/孔洞概率的量化收益），难以重复实施。

四、新颖性与创造性初步评价

• 同轴多传感（光学/声发射）采集、以时序深度学习进行焊接质量判别、并基于判别结果闭环调节激光功率/脉冲参数的技术思想，为激光焊接质量控制领域的常见路线。业界已有将光学发射、等离子体辐射或声学信号用于缺陷预测与功率闭环的公开，亦有采用一维CNN或其他深度网络进行实时分类的报道。基于此，C1+C2+C3的简单并列组合存在显著的创造性挑战。
• 可能的可专利性切入点：
  • 将“孔洞风险预测”与“跳点（轨迹重排序/时间间隔重分配）”形成紧耦合控制，并给出可验证的热管理/缺陷概率下降的量化规则，若现有技术多停留在功率调节而非轨迹在线重排，则有机会形成区别性技术效果。
  • 对“≤2 ms”低时延闭环在设备级实现的具体技术手段（传感器硬件耦合、光路设计、数据通路、模型量化与流水线并行、控制器设计）的限定，若能证明相较已知系统实现更高的闭环频宽且显著降低虚焊/孔洞率，亦可能支撑创造性。
  • “轨迹哈希+特征向量”的追溯若与在线自检、防篡改、以及后续重焊参数的自动决策闭环形成可验证的工艺质量改进链路，可提升技术性并弱化“呈现信息”之评价风险。
• 建议在检索与比对后，围绕上述差异化特征收窄权利要求范围，并提供对比试验数据或统计显著性，支撑技术效果。

五、权利要求修改与补充建议 为提升清楚性、支持性与创造性，建议：

• 对独立权利要求（方法）：
  • 明确I(t)为特定波段（例如400–900 nm）等离子体辐射/反射光强，S(t)为X–Y kHz带宽的声发射信号；限定同轴集成的分光/滤光组件与传感器安装几何。
  • 限定滑动窗口长度、步长、CNN层级/卷积核大小上限、参数量或运算预算（MACs），以及分类标签集合与决策阈值。
  • 增加“熔深估算模块”，限定经标定得到的d̂=f(I,S)或分类到d区间的映射方法；给出标定流程的必要技术特征（试样、厚度范围、拟合误差上限）。
  • 明确控制律（例如：以d̂与[dmin,dmax]的偏差，经增益受限的PI调节E与D；更新周期≤1 ms；输出变化率限幅），并限定执行器与通信延迟预算以满足“≤2 ms”。
  • 对“跳点策略”进行过程性限定：风险评分超过阈值θ时，跳过当前N个焊点并插入冷却间隔Δt，或重排至空间距离≥L的下一点；并在回程时采用降能预热/多脉冲策略复焊。
  • 将追溯数据的技术用途写入方法流程：用于在线参数自适应或终检自动放行/拦截逻辑，而非单纯存储/展示。
• 设置关键从属权利要求：
  • 具体网络结构与量化/蒸馏/剪枝方案；窗口长度与频域增强（STFT或小波）的可选实施例。
  • 传感硬件频响/灵敏度与抗干扰手段；I与S的多模态融合策略（早期融合/后期融合）。
  • 轨迹重排序的约束（热累积模型、相邻点最小间距、产线节拍影响上限）。
  • 设备权利要求：焊接头、同轴光路、前端采集、边缘计算单元与执行控制器的结构与连接关系，保证方法权利要求与装置权利要求相互支撑。
• 删除或限定容易被认为为非技术性的要素表述（例如“二维码”本身），改以“以机器可读标识写入并用于制造执行系统触发控制策略”的技术性描述。

六、说明书应补充的技术内容

• 传感链路：光路结构图、带宽与滤光、声发射安装位置、标定与抗噪策略。
• 数据处理：采样同步、时钟精度、ADC位宽、前端滤波、数据打包与DMA。
• 模型训练：数据规模、标注方法（熔深/孔洞的检测基准）、训练/验证/测试集划分、域适配（不同材料/厚度/设备）、实时推理加速方法（定点化、流水线、批量=1）。
• 控制律与稳定性：离散化控制模型、参数范围、相位裕度/增益裕度或仿真/实测响应，确保在2 ms内完成“采集-推理-决策-执行”闭环。
• 性能数据：对比无闭环/仅功率闭环/加入跳点策略的缺陷率、返修率、熔深偏差统计及显著性检验。
• 适用边界：材料（铝/铜/镍等）与厚度范围、表面状态、激光器类型（脉冲/连续、光斑直径）、不可用场景与失效模式。

七、国内外可专利性与策略提示

• 中国（CN）：制造过程控制方法具备客体适格性。应避免以“数据呈现/业务规则”为核心的从属项。建议以设备+方法双独立权利要求，同步提交对比实验数据以支撑创造性与技术效果。
• 欧洲（EP）：EPO对混合发明中非技术特征的贡献不计入创造性评估。建议将“跳点策略引发的热管理改善”和“低时延闭环实现细节”作为解题手段，避免追溯数据的非技术性权利要求。权利要求需满足Art.84清楚性、Art.83实施可行性。
• 美国（US）：制造方法通常可通过§101审查，但与“数据记录/二维码/绑定序列号”相关的抽象要素应与具体机器控制步骤紧密耦合。§112对“轻量”“自适应”等术语要求明确界定与实施例支撑。
• 日本（JP）：对工业过程控制与AI应用较为友好，但同样强调具体技术手段与可验证效果。建议将“跳点策略+闭环功率调节”的协同效应以实验数据支撑。

八、检索与对比建议（方向性） 为论证创造性，建议重点检索并区分以下技术路线：

• 激光焊接同轴传感（光学/声学/光学同轴测量）+功率闭环的商用系统与文献；
• 使用一维CNN/时序深度网络进行焊接质量在线判别的研究/专利；
• 电池极耳焊接过程的“缺陷预测后路径/节拍在线重排（跳点/延时）”相关公开（该点或为本案差异化来源）。

九、结论与办理意见

• 请对权利要求进行实质性收窄与术语明确化，补充说明书以满足第26条第3、4款要求。
• 在补充分明技术手段、可复现实施条件及性能数据后，再行评估创造性。当前文本下，权利要求1存在清楚性与支持性瑕疵，且创造性尚未得到充分证明；权利要求2存在非技术性风险。建议在国内首次申请递交前完成上述补正与数据准备，以利后续PCT及各局审查阶段的稳定性与可授权性。

Examiner’s evaluation of patentability and formal compliance

Application title: Heterogeneous database incremental synchronization using graph-based schema mapping and lock-free conflict resolution

Technical field: Data synchronization for OLTP-to-analytics pipelines

Summary of independent claim 1 (as understood)

• G1: Construct a property graph where nodes represent source/target fields and edges encode mapping functions with version tags.
• G2: Capture change streams via CDC and assign vector clocks per key.
• G3: Resolve conflicts without locks by using CRDT-style merges for commutative fields and per-key deterministic replay for non-commutative transactions.
• G4: Preserve ACID at the target by micro-batching idempotent upserts with write-ahead checksums and deduplication indices.
• Additional limitation: Vector clocks are compacted using probabilistic sketches to bound metadata size per partition.

Claim 2 adds schema drift detection with automatic mapping update proposals and human-in-the-loop approval.

1. Subject-matter eligibility

United States (35 U.S.C. § 101; MPEP § 2106)

• The claims are directed to a “process,” a statutory category. The core concept involves data synchronization and conflict resolution, which, in the abstract, can be characterized as information processing. However, there are specific recitations of technical data structures and execution mechanisms (CDC streams, vector clocks, CRDT merges, write-ahead checksums, deduplication indices, and probabilistic sketches).
• Step 2A(Prong Two): The recited elements, if sufficiently specified, appear integrated into a practical application that improves computer functionality in distributed data consistency, throughput, and metadata overhead. The probabilistic compaction of vector clocks, if concretely taught to preserve causal comparison with bounded error and defined fallbacks, weighs in favor of eligibility.
• Step 2B: If the specification demonstrates that the combination yields a specific technical improvement (e.g., bounded metadata per partition while maintaining correct causal/conflict decisions with defined error handling) beyond generic computer implementation, the claims are likely patent-eligible. As drafted, functional language (e.g., “preserve ACID,” “lock-free conflict resolution”) risks being characterized as results-oriented. Amendments providing concrete algorithms and data structure interactions are recommended to avoid § 101 vulnerabilities.

Europe (Art. 52(1) EPC)

• The claims address a technical problem (consistent and efficient synchronization among heterogeneous databases) and recite technical means. Provided the specification discloses concrete algorithmic details for the vector-clock compaction and conflict resolution mechanisms, the claimed subject-matter would likely have technical character and be eligible under Art. 52 EPC. Section 2 below addresses inventive step.

2. Novelty and inventive step

General observations (without a prior art search)

• Elements similar to G1–G4 are commonly found in distributed systems: graph-based schema mappings with versioning, CDC pipelines, vector clocks for causality, CRDT-based resolution for commutative updates, deterministic replay for non-commutative operations, idempotent upserts with deduplication, and WAL-based durability are widely known and used.
• The distinguishing feature appears to be the compaction of vector clocks using probabilistic sketches to bound metadata size per partition. The integration of a versioned property-graph schema mapping with CDC and conflict resolution may also be asserted as a cohesive architecture, but absent evidence of a synergistic effect, such aggregation can be found obvious where each component is used for its known purpose.

Risk of obviousness (35 U.S.C. § 103; Art. 56 EPC)

• Absent specific algorithmic detail, “vector clock compaction using probabilistic sketches” risks being viewed as a routine application of known sketching techniques (e.g., Bloom filters, count-min) to a known metadata-scaling problem. Moreover, if the compaction does not preserve the partial order properties necessary for causality determination—or if it does so only with unspecified error properties—it may be deemed an obvious, and possibly defective, approximation. The inventive step hinges on:
  • A concrete compaction method that enables sound or bounded-error causality comparisons;
  • A defined decision procedure for concurrency vs. causality under the sketch, with quantifiable false positive/negative trade-offs and a specified fallback mechanism;
  • Integration that yields an unexpected technical effect (e.g., strictly bounded per-partition overhead with maintained correctness at specified confidence, leading to improved throughput under bursty loads without locking).
• Claim 2’s schema-drift detection and update proposals are, in general, known in ETL tooling and data integration systems. Without specific detection algorithms (e.g., constraint-based or statistical schema inference) and a concrete feedback/control loop tied to the property graph and CDC pipeline, this feature is likely obvious.

Recommendation: To support novelty and non-obviousness, include detailed algorithms, complexity bounds, probabilistic guarantees, and empirical evidence showing a counterintuitive or non-linear improvement over conventional techniques (e.g., dotted version vectors, interval tree clocks, or metadata sharding). Provide clear reasons why a skilled person would not straightforwardly apply generic sketches to vector clocks due to causality-comparison challenges, and how your approach overcomes these challenges.

3. Clarity and definiteness

United States (35 U.S.C. § 112(b); MPEP § 2173)

• Indefinite or result-oriented terms needing definition or algorithmic specificity:
  • “Property graph” (define node/edge schema, allowed properties, and mapping function semantics).
  • “Edges encode mapping functions with version tags” (specify function types, invertibility, versioning and compatibility rules).
  • “Vector clocks are compacted using probabilistic sketches” (identify sketch type(s), update rules, comparison procedure, error bounds, and fallback).
  • “Lock-free conflict resolution” (specify the concurrency progress guarantee, scope—per-key or global—and the non-blocking algorithm used).
  • “CRDT-style merges for commutative fields” (identify CRDT types supported, semilattice definitions, and merge laws).
  • “Per-key deterministic replay for non-commutative transactions” (define the ordering rule, e.g., vector-clock partial order with deterministic tie-breakers).
  • “Preserve ACID at target by micro-batching idempotent upserts with write-ahead checksums and deduplication indices” (define the micro-batch commit protocol, isolation model, dedup key structure, WAL checksum algorithm, and recovery procedure).
  • “Bound metadata size per partition” (quantify bound).
• The current claim language is heavily functional, asserting outcomes without reciting the technical means with sufficient precision. This may give rise to § 112(b) indefiniteness and, in Europe, an Art. 84 EPC clarity objection.

Europe (Art. 84 EPC; Guidelines F‑IV)

• Similar clarity issues arise, particularly with open-ended functional terms and absence of algorithmic parameters essential for understanding the scope.

4. Enablement and written description

United States (35 U.S.C. § 112(a); MPEP § 2164, § 2163)

• Enablement risk: The application must teach how to implement the probabilistic-sketch compaction of vector clocks such that causality and concurrency decisions are made correctly or within specified error tolerances, including remediation steps. Without a concrete algorithm, a skilled person cannot practice the full scope without undue experimentation.
• Written description: Generic references to “CRDT-style merges,” “lock-free conflict resolution,” and “idempotent upserts” do not demonstrate possession of a specific implementation across the breadth of “heterogeneous databases.” The specification should provide exemplars and generalization principles.
• Scope vs. enablement: Claim 1 is not limited to particular databases, CRDTs, or sketch types. If only certain combinations are practicable, claims should be commensurate with the embodiments.

Europe (Guidelines F‑III, C‑II)

• The same concerns apply regarding sufficiency of disclosure, particularly for the compaction of vector clocks and ensuring ACID semantics at the target with micro-batching.

5. Unity of invention

• The claims appear to share a single general inventive concept (incremental synchronization with the particular compaction/conflict resolution scheme). Unity is likely satisfied.

6. Suggested amendments and additions

To address §§ 101/112 and Art. 84/56 issues, consider amending as follows:

• Define the property-graph mapping:
  • Nodes: typed fields with unique identifiers; edges: typed mapping functions f: S → T with invertibility flags; version tags include semantic version and compatibility metadata; a constraint that evaluation of mappings is referentially transparent for CDC events.
• Specify CDC and vector clocks:
  • Per-partition logical time base; per-key vector clock maintained as a sparse map keyed by partition identifiers.
• Probabilistic compaction of vector clocks:
  • Identify the sketch (e.g., count-min sketch with conservative update or a Bloom filter vector plus per-partition counters).
  • Describe update rule for increments, and a compare(a, b) procedure that decides causality, concurrency, or indeterminate with bounded false rates ε.
  • Provide fallback to a materialized vector for indeterminate cases, with cache TTL and eviction policy, thereby preserving correctness.
  • State explicit bounds on metadata per partition (e.g., O(1) words) and quantify error probabilities and their effect on conflict resolution.
• Lock-free conflict resolution:
  • Restrict to per-key lock-free CAS loops with retry, or wait-free bounded retries, and state progress guarantees.
• CRDT merges and deterministic replay:
  • Enumerate supported CRDT types (e.g., LWW-register with hybrid logical clocks; G-Counter; PN-Counter; OR-Set; delta-state CRDTs) and provide algebraic merge functions.
  • For non-commutative ops, define deterministic ordering: primary key; then causal order; ties broken by a total order on (partition-id, sequence, hash).
• ACID preservation at target:
  • Define micro-batch boundaries; idempotency keys (e.g., H(key || op-payload || clock-digest)); dedup index structure (e.g., LSM-tree keyed by idempotency key with tombstones).
  • WAL checksum computation and recovery workflow; isolation mode (e.g., snapshot isolation with write-write conflict detection at commit).
  • Commit protocol with pre-commit verification of dedup entries and atomic publish of batch.
• Schema drift detection and mapping update (claim 2):
  • Define automatic detection rules (e.g., structural diffs, constraint violations, statistical type inference).
  • Generate update proposals by graph rewrite rules with version bump and compatibility proof obligations; deploy via staged shadow evaluation and human approval; roll-back policy.

7. Possible dependent claims to strengthen patentability

• The sketch is a count-min sketch with conservative update and width/ depth parameters chosen to bound comparison error below ε; include a theorem or lemma on causality decision error.
• A fallback cache of exact vector clocks is maintained for keys whose sketch comparison is indeterminate, with TTL τ and adaptive resizing based on observed drift rate.
• The deduplication index is keyed by a digest of (entity-id, canonicalized operation descriptor, compacted-clock digest) and is implemented as a lock-free concurrent hash-trie.
• The CRDT merge function for OR-Set uses delta-state encoding with causal context derived from the compacted clock plus fallback reconciliation.
• The micro-batch commit uses a two-phase barrier ensuring all idempotent upserts are durable in WAL with checksums before visibility switch under snapshot isolation.

8. Evidence and technical effects

• Include experimental data showing:
  • Bounded metadata per partition independent of key cardinality.
  • Throughput/latency improvements under bursty loads vs. a baseline without compaction.
  • Observed rates of sketch-induced indeterminacy and the effectiveness of the fallback cache.
  • Preservation of ACID at the target (no lost updates, correct isolation) verified by fault-injection tests.

9. Conclusion

• Eligibility: Likely patent-eligible if claims are amended to recite concrete technical means and improvements.
• Novelty/inventive step: As drafted, each of G1–G4 aligns with well-known techniques; novelty likely resides in the probabilistic compaction of vector clocks and its integration with conflict resolution and micro-batched ACID upserts. Without concrete algorithms, error bounds, and a proven technical advantage, an obviousness rejection is likely.
• Clarity/enablement: Present claims are predominantly results-oriented and lack algorithmic detail, creating substantial § 112(a)/(b) and Art. 84/EPC risks.

Actionable guidance: Amend independent claim 1 to recite specific data structures, algorithms, parameters, comparison procedures, and commit protocols. Limit claim scope to the disclosed embodiments that demonstrably achieve the asserted technical effects. Enhance the specification to provide enabling detail for the probabilistic compaction of vector clocks, the lock-free conflict resolution mechanism, and the ACID-preserving micro-batch commit protocol.