Patent Mining and Portfolio Layout in the AI Era

AI Turns Patent Mining into a Structured Combination Game

In traditional patent mining, technical disclosure often arrives like toothpaste squeezed from a tube: inventors write whatever they happen to remember, while patent engineers search scattered keywords and worry about missing prior art. In complex technical systems, this method easily leaves gaps.

AI changes the workflow. It does not replace human creativity or legal judgment, but it is excellent at structuring information, expanding candidate features, and running large combination exercises. The practical method is a table-based approach.

AI Turns Patent Mining into a Structured Combination Game: practical detail

For AI-era R&D teams, patent engineers, and technology managers, this point should be treated as a working step rather than a slogan. The practical work is to connect the article's idea with functional decomposition, AI-assisted technical-means expansion, cross-field substitution, matrix combination, conflict filtering, and expert screening. That is what turns a general insight into a repeatable professional service.

The team should record concrete evidence: function-feature tables, AI prompt records, candidate technical means, feasibility notes, prior-art overlaps, and portfolio value scores. Without this evidence layer, the method remains only an opinion. With it, the article becomes useful for client communication, internal decision-making, patent drafting, prosecution strategy, and later portfolio review.

A complete application of this section normally ends with a decision: which AI-generated combinations are technically credible, commercially meaningful, and legally supportable enough to become patent applications. Ma Su's examiner background matters here because the decision is not based only on enthusiasm; it is tested against technical contribution, support in the disclosure, likely examination reasoning, and business value.

Step One: Decompose the Existing Technical System

Take a power battery mechanical structure as an example. The team can decompose it into functional positions: housing structure, cooling structure, support structure, sealing structure, fastening method, thermal insulation, crash protection, monitoring layout, and manufacturing process.

Each row of the table represents a technical function. The current technical feature is then placed into that row. AI can assist by reading patents and papers, suggesting missing functional positions, and making the decomposition more complete.

Step One: Decompose the Existing Technical System: practical detail

For AI-era R&D teams, patent engineers, and technology managers, this point should be treated as a working step rather than a slogan. The practical work is to connect the article's idea with functional decomposition, AI-assisted technical-means expansion, cross-field substitution, matrix combination, conflict filtering, and expert screening. That is what turns a general insight into a repeatable professional service.

The team should record concrete evidence: function-feature tables, AI prompt records, candidate technical means, feasibility notes, prior-art overlaps, and portfolio value scores. Without this evidence layer, the method remains only an opinion. With it, the article becomes useful for client communication, internal decision-making, patent drafting, prosecution strategy, and later portfolio review.

A complete application of this section normally ends with a decision: which AI-generated combinations are technically credible, commercially meaningful, and legally supportable enough to become patent applications. Ma Su's examiner background matters here because the decision is not based only on enthusiasm; it is tested against technical contribution, support in the disclosure, likely examination reasoning, and business value.

Step Two: Replace Features Across Fields

For each functional position, ask AI to list all technical means that can perform the same function, regardless of industry. A cooling position may generate liquid cooling plates, side cooling, immersion cooling, heat pipes, forced air cooling, phase-change materials, or direct refrigerant cooling. A support position may generate honeycomb structures, foam metal, truss structures, integrated cast ribs, or composite fillers.

The value of AI is breadth. Human engineers tend to remain inside their own industry. AI can help bring in structures from aerospace, construction, electronics, biomimetics, and chemical engineering.

Step Two: Replace Features Across Fields: practical detail

For AI-era R&D teams, patent engineers, and technology managers, this point should be treated as a working step rather than a slogan. The practical work is to connect the article's idea with functional decomposition, AI-assisted technical-means expansion, cross-field substitution, matrix combination, conflict filtering, and expert screening. That is what turns a general insight into a repeatable professional service.

The team should record concrete evidence: function-feature tables, AI prompt records, candidate technical means, feasibility notes, prior-art overlaps, and portfolio value scores. Without this evidence layer, the method remains only an opinion. With it, the article becomes useful for client communication, internal decision-making, patent drafting, prosecution strategy, and later portfolio review.

A complete application of this section normally ends with a decision: which AI-generated combinations are technically credible, commercially meaningful, and legally supportable enough to become patent applications. Ma Su's examiner background matters here because the decision is not based only on enthusiasm; it is tested against technical contribution, support in the disclosure, likely examination reasoning, and business value.

Step Three: Combine, Filter, and Review

Once each row has candidate features, complete technical schemes can be generated by combining one option from each row. This may create thousands of possibilities. AI can quickly remove combinations with physical conflict, obvious impracticality, or clear prior-art overlap.

The remaining list is not yet a patent portfolio. It is a seed library. Product managers, engineers, and patent professionals must evaluate cost, manufacturability, market direction, grant probability, and defensive value.

Step Three: Combine, Filter, and Review: practical detail

For AI-era R&D teams, patent engineers, and technology managers, this point should be treated as a working step rather than a slogan. The practical work is to connect the article's idea with functional decomposition, AI-assisted technical-means expansion, cross-field substitution, matrix combination, conflict filtering, and expert screening. That is what turns a general insight into a repeatable professional service.

The team should record concrete evidence: function-feature tables, AI prompt records, candidate technical means, feasibility notes, prior-art overlaps, and portfolio value scores. Without this evidence layer, the method remains only an opinion. With it, the article becomes useful for client communication, internal decision-making, patent drafting, prosecution strategy, and later portfolio review.

A complete application of this section normally ends with a decision: which AI-generated combinations are technically credible, commercially meaningful, and legally supportable enough to become patent applications. Ma Su's examiner background matters here because the decision is not based only on enthusiasm; it is tested against technical contribution, support in the disclosure, likely examination reasoning, and business value.

The Human Role Becomes More Important

AI can fill the table, but humans decide what matters. A carbon-fiber housing with side cooling may be too expensive for current customers but valuable as a defensive patent. A composite housing with integrated support may match a next-generation product. A simple fastening substitution may be novel but unreliable.

In the AI era, patent mining becomes less like waiting for inspiration and more like structured innovation design: functional decomposition, AI expansion, matrix combination, expert screening, and portfolio decision.

The Human Role Becomes More Important: practical detail

For AI-era R&D teams, patent engineers, and technology managers, this point should be treated as a working step rather than a slogan. The practical work is to connect the article's idea with functional decomposition, AI-assisted technical-means expansion, cross-field substitution, matrix combination, conflict filtering, and expert screening. That is what turns a general insight into a repeatable professional service.

The team should record concrete evidence: function-feature tables, AI prompt records, candidate technical means, feasibility notes, prior-art overlaps, and portfolio value scores. Without this evidence layer, the method remains only an opinion. With it, the article becomes useful for client communication, internal decision-making, patent drafting, prosecution strategy, and later portfolio review.

A complete application of this section normally ends with a decision: which AI-generated combinations are technically credible, commercially meaningful, and legally supportable enough to become patent applications. Ma Su's examiner background matters here because the decision is not based only on enthusiasm; it is tested against technical contribution, support in the disclosure, likely examination reasoning, and business value.

Original Chinese Essay

专利还能如此简单？AI时代的专利挖掘与布局

AI时代如何做专利挖掘？试试这张“表格”的无限组合游戏

在传统的专利布局工作中，我们常常面临一个困境：技术交底书像挤牙膏，发明人想到哪写到哪；专利工程师检索时像在大海捞针，总担心有漏网之鱼。

尤其是在技术密集型的领域，当我们要对一个复杂的系统（比如机械结构部分）进行全面布局时，单靠头脑风暴和零散的灵感，很容易留下专利壁垒的空缺。

现在，AI的出现正在改变这一局面。它不能替代人类的创造性思维，但可以成为我们思维的“放大器”。今天，我想分享一种AI辅助下的专利挖掘方法——表格法。这其实是一场关于技术特征的“无限组合游戏”。

第一步：解构与拆解，把技术放进表格里

AI最擅长的是什么？是处理结构化的信息。

当我们想要对动力电池的机械结构进行全面布局时，首先要做的不是想“还能做成什么样”，而是把“现在是什么样”进行功能性的极细拆解。

假设我们有一个典型的电池包，我们可以将其机械结构拆分为：

· 位置A（箱体结构）：铝合金托盘

· 位置B（冷却结构）：底部水冷板

· 位置C（支撑结构）：内部横梁

· 位置D（密封结构）：橡胶密封条

· 位置E（固定方式）：螺栓连接

......

我们把现有的技术方案拆解开来，填入表格的行中。每一个位置，都代表一种技术功能。而对应的内容，则是实现该功能的具体技术特征。

这一步，AI可以通过对现有专利、论文的阅读，辅助我们进行更标准、更细致的功能分区，确保我们的“拆解清单”没有遗漏。

第二步：特征替换，AI的穷举狂欢

表格画好后，真正的游戏开始了。

对于表格中的每一个“位置”（技术功能），我们向AI提问：

“请列出所有能够实现‘底部冷却’功能的技术特征，不论领域，不论形态。例如，可以是液冷管、可以是相变材料、可以是风冷翅片，也可以是半导体制冷片……”

这就是全领域搜索。

人类的思维容易受到惯性限制，做电池的就只盯着电池行业。但AI的知识库横跨机械、化工、电子、建筑甚至仿生学。它可以为我们每一个功能位置，生成一个长长的、甚至有些天马行空的“候选特征清单”。

· 位置A（箱体）：铝合金 → {碳纤维复合材料、高强度钢、注塑件、泡沫铝……}

· 位置B（冷却）：底部水冷板 → {侧面冷却、浸没式冷却、热管、强制风冷、制冷剂直冷……}

· 位置C（支撑）：内部横梁 → {蜂窝铝板、泡沫金属填充、桁架结构、一体铸造成型筋……}

通过AI，我们完成了第一轮“暴力枚举”，把原本空白的表格，填得满满当当。

第三步：排列组合，AI的快速筛选

现在，我们有了一个二维矩阵：

· 行（位置）： A, B, C, D, E…

· 列（选项）： A1, A2, A3… ; B1, B2, B3… ; C1, C2, C3…

理论上，一个完整的动力电池机械结构方案，就是从这个矩阵的每一行中选出一个选项，进行组合。这会产生成千上万种新方案。让人类逐一去构思，是不现实的。

这时，又轮到AI登场。我们可以设定指令：

“请将上述各技术位置的候选特征进行随机组合，生成1000个完整的技术方案。然后，根据以下标准进行初步筛选：

1. 结构是否存在物理冲突？（例如，选择了浸没式冷却就不能同时选择不防液的箱体材料）

2. 从现有技术文献推断，是否具备新颖性？（即这个组合在现有专利中没出现过）

3. 初步判断是否具备基本的工业实用性？

AI会在极短时间内，将这1000个组合缩减到几十个“逻辑上可行”且“可能具备专利性”的方案。

第四步：人类智慧，最终的价值判断

经过AI的快速筛选，我们拿到了一份经过初步清洗的“专利种子清单”。

但这还不是终点。AI不知道成本，不知道市场趋势，不知道生产工艺的难度。这些方案最终要交到产品经理团队的手里。

产品经理们会从商业和工程的角度进行审视：

· 方案X：用了碳纤维箱体+侧面冷却，虽然轻，但成本太高，客户不买单。但这是一个极具威慑力的“防御性专利”，可以占坑。

· 方案Y：用了复合材料箱体+一体化桁架支撑，减重效果明显，成本增加可控，符合下一代高端车型的需求。这可以作为核心专利进行重点布局。

· 方案Z：单纯把螺栓连接换成了卡扣连接，虽然新颖，但可靠性存疑。先作为储备方案搁置。

最终，人类的商业洞察力，为这场由AI驱动的技术组合游戏，画上了句号。

在AI时代，专利挖掘不再是一场漫无目的的搜寻，也不是一次灵光一闪的发明。它更像是一场结构化的创新设计。

通过“功能拆解 → AI穷举 → 矩阵组合 → 人工审视”的表格法，我们能够：

1. 拓宽视野：打破行业壁垒，引入跨领域技术。

2. 查漏补缺：系统性地覆盖所有技术路径，不给竞争对手留下空隙。

3. 提升效率：让AI完成繁重的数据检索和初步组合工作，让人类专注于高价值的决策。

下一次当你面对一个复杂的技术模块时，不妨打开一个空白表格，让AI做你的“参谋”，帮你填满格子，然后开始玩这场关于未来的组合游戏吧。