An Integrated Method to Acquire Technological Evolution Potential to Stimulate Innovative Product Design

Abstract

Fast and effective forecasting of the new generation of products is key to enhancing the competitiveness of a company in the market. Although the technological evolution laws in the theory of the solution of inventive problems (TRIZ) have been used to predict the potential states of products for innovation, there is a lack of effective methods to select the best technological evolution law consistently with product replacement and update, and acquiring potentially new technologies and solutions, which relies heavily on designers’ experience and makes it impossible for designers to efficiently use the technological evolution laws to stimulate product innovation. Aimed to bridge this gap, this paper proposes an integrated method consisting of three main steps, combining the technological evolution laws with back propagation neural network (BPNN), international patent classification (IPC) knowledge and company’s technological distance. The best technical evolution law is first searched by a BPNN. The functional verbs and effects in the IPC are then extracted and searched for potential technologies in the Spyder-integrated development environment. Finally, the company’s technological distance is used to select analogous sources of potential solutions in the patent database. The final innovative design is determined based on the ideality. The proposed method is applied in the development of a steel pipe-cutting machine to verify its feasibility. The proposed method reduces the dependence on designers’ experience and provides a way to access cross-domain technologies, providing a systematic approach for the technological evolution laws to motivate innovative product design.

1. Introduction

1.1. General Background

Innovative products can maintain a company’s competitiveness in response to rapid market change [1,2]. The innovative product requires the creation of new design concepts [3], it is essential to have an effective method in the development of an innovative product concept and corresponding design solution.

Customers may lack foresight and have no idea of future product requirements [4,5]. A company needs to take its own initiative and proposes new technologies for product innovation, as a basis for guiding customer requirements [6,7]. Product development requires accurately analyzing the technical level of existing products to predict their new technologies [8,9]. Altshuller identified technological evolution trends for technological evolution laws by analyzing patent data [10]. It was found that the technological evolution laws in one engineering field can be applied in other fields [11,12]. The principle is to determine the potential state of the product from the current technology [13].

Comparing the classical technological forecasting methods, such as morphological analysis [14] and the Delphi method [15], technological evolution laws in the theory of the solution of inventive problems (TRIZ) provide effective tools to identify the potential of products [16]. Research has been conducted on the technological evolution laws for theoretical development [17,18,19] and practical applications [20,21], but there are still problems:

• Selecting the best technological evolution law consistent with the product replacement and update, predicting the potential feature of a product, is subjectively influenced by the designer.
• It relies heavily on the designer’s experience in selecting potential technologies for the features of the product.
• Potential solutions for potential technologies usually require a cross-domain search for analogy. Self-built databases affect the analogy results due to the limited case sources. Online database searches do not consider the feasibility of the technology transfer.

1.2. Technological Evolution Laws

The technological evolution laws provide tools to study the technological development trends of an existing product [22,23,24] and its next evolution state for product improvement [25,26]. Research on the technological evolution laws encompasses both theoretical and product developments. Cascini [18] proposed an algorithm for functional analysis to build an evolution network with repeatable steps. Shpakovsky [27] built an evolution tree to analyze product development trends for technological gaps, it remains the application of the technological evolution laws. Sheu [28] proposed 52 technological evolution laws based on the study of Mann [29] and used mathematical algorithms to select the technological evolution laws. Analogous solutions were searched in a self-built case database, limited by the number of cases already input. Li [30] combined technological evolution with trimming [31] to select technological opportunities for new designs based on existing bundled patent portfolios. With the development of computer technologies, applications of the technological evolution laws have been aided by computers. Zhang [32] developed a prediction system for the technology evolution potential state based on a radar diagram of the product evolution. Yoon [33] and Park [34] extracted patent information. By comparing the relationships between the extracted information and technological evolution laws, the technical state of the product could be determined. This approach was a procedural implementation of the evolution tree and lacks an effective way for identifying potential technologies and solutions. Wang [35] built a back propagation neural network (BPNN) [36] by collecting sample cases. The corresponding technological evolution law can be decided from engineering parameters [37]. The training process did not consider the order of the engineering parameters, which is limited to predict the technological evolution law based on the causes of problems that have already occurred.

Technological evolution laws have been applied in different fields. Yang [38] fused the technological evolution laws and product life cycle in designing a new eco-mobile phone. Vidal [39] established the relationship between fuzzy cognitive maps and technological evolution laws applied in Spanish ceramics. Rahim [40] combined the technological evolution laws with chemical engineering in the development of materials for automobile bumpers. Wang [41] combined the technological evolution laws with a technological maturity mapping system to analyze the evolution of waste gas for recycling printed circuit boards.

However, existing research has paid little attention to the selection of the best technological evolution law corresponding to product replacement and update, and an effective way of acquiring potentially new technologies and solutions. In order to improve the existing states, this paper builds a new BPNN model based on the previous research of our team [35]. In different life cycle stages, the same engineering parameters may be extracted, but they will correspond to different technological evolution laws. Therefore, the training process is improved for the selection of the best technological evolution law. At the same time, the international patent classification (IPC) [42] knowledge and company’s technology distance [43] are introduced. Functions [44] and effects [45] co-occurring in the IPC are used to acquire potentially new technologies. The company’s technology distance is used for the selection of analogous sources of potential solutions.

1.3. Acquisition of IPC Knowledge

The IPC was proposed according to the Strasbourg Agreement as an internationally used tool for classifying and searching patent documents [46]. The patent classification system was established based on a hierarchical structure, including the section, class, subclass, main group and subgroup [47,48].

The IPC can be used not only to search the technological domain information [49], but also acquire design knowledge [50]. Corresponding technological topics under each entry include specific design information. Technological features in each subclass and following groups are usually related to the function and effect, which can be extracted as the design knowledge [51]. According to the discovery and reorganization of the knowledge associations in the IPC, Qiu [52] used the extracted knowledge in the function innovation of a product. Ji [53] provided a method for the exchange of the function and effect knowledge by relevancy or universality based on the number of co-occurrences of a functional verb or effect in each category in the IPC. However, it ignored the filtering method of irrelevant functional verbs or effects during the initial screening, and composition rules of the combinatorial effect. Using a coin-sorting device as an example of the comparison experiment, it was found that the group with IPC knowledge as a stimulus had a higher innovation output than the other groups, which shows a significant effect on the knowledge stimulation [54].

The potential technology is based on the improvement of the function or effect. The function and effect knowledge contained within the IPC can be used as an effective knowledge stimulation source. Natural language processing (NLP) was used in this study [55]. Based on the natural language technology platform (LTP) for Chinese text, developed by Harbin Institute of Technology [56], and improved effect composition rules, functional verbs and effects in the IPC were extracted in a Spyder-integrated development environment, which was used as a database to search potentially new technologies.

1.4. Acquisition of a Patent Technology Database on the Company’s Technological Distance

Product development relies heavily on previous design experience and knowledge rather than building completely new products from scratch [57,58]. Effective knowledge stimulation can facilitate product idea generation and enhance design quality [59]. Previous cases with a similar function or principle are used to design a new product [60,61,62]. The patent database contains rich design knowledge from various fields. According to a survey by the World Intellectual Property Organization, 90–95% of invention and creation in the world is issued in the form of a patent. Patent knowledge can be effectively used to reduce product development time by an average of 60% [63]. Patent databases with large amounts of technological information present opportunities for designers, but inappropriate selection of technologies can affect the innovation output. For example, new energy vehicles can more easily absorb and utilize technologies from the company that produces batteries than a chemical company that research battery materials [64].

A company’s technological distance is the degree of differentiation between partners’ technological knowledge areas and the company’s current active areas [65]. Several scholars have found an inverted U-shaped relationship between innovation output and the company’s technological distance [66,67]. If the company’s technological distance is high, the knowledge overlap between the companies is low, and the target company will have difficulty in digesting and absorbing the technology. If the overlap of knowledge is high, there is a high similarity of technology stock between the companies, which is not conducive to the generation of novel technologies [68,69].

Potential technologies obtained by the technological evolution laws are usually not involved in the company or field application, which requires cross-domain knowledge acquisition. Resource complementarity and the company’s technological distance provide an effective way for cross-domain knowledge acquisition [70,71]. Therefore, this research uses the company’s technological distance to select analogous sources of potential solutions, different from existing studies that use the company’s technological distance in the analysis of alliance companies.

1.5. Summary of the Literature

In summary, for deficiencies in the technological evolution laws in the process of product innovation, this paper proposes an integrated method to obtain the technological evolution laws, potential technologies and potential solutions. Advantages of this method are as follows.

• BPNN provides a tool for the selection of technological evolution laws corresponding to product replacement and update, reducing the subjective influence of the designer.
• The knowledge of function and effect in the IPC is extracted as a database, providing a source of motivation for the acquisition of potential technologies and reducing the dependence on design experience.
• The company’s technology distance provides a measure to select analogous sources for potential solutions from a patent database of the best analogous sources and cross-domain technologies.

1.6. Document Organization

The following parts of this paper are organized as follows. Section 2 describes the methods of the technological evolution, IPC knowledge, and a company’s technological distance. Section 3 uses a case study to explain the application process and verify the feasibility of the proposed method. Section 4 discusses the research and proposes future work. Section 5 concludes the research.

2. Proposed Method

2.1. Analyzing the Potential Features of a Product

A complete product system is composed of energy, transmission, actuation and control devices [16]. In order to facilitate the analysis of the trends in product replacement and update, the four subsystems mentioned above were analyzed as the target objects in this research.

To select the best technological evolution law consistent with product replacement and update, a new BPNN was trained based on the research of our team [35]. Engineering parameters were extracted for different stages of product replacement and update. The extracted engineering parameters and their order of appearance were used as the input for training the BPNN, the corresponding technological evolution law was used as the output. 100 cases were randomly collected from patent databases and literature as training and testing sets for the BPNN at a ratio of 4:1. Three domain experts with skills in the technological evolution laws first jointly analyzed the engineering parameters corresponding to the update process of each case and their compliance with the technological evolution laws. The engineering parameters, order and technological evolution laws corresponding to each case were then used as the input and output of the BPNN. The trained model was saved for the application. When a designer uses it, the engineering parameters and order are the input of the saved model to predict the corresponding technological evolution law. The BPNN code was written and ran in the MATLAB software. The visualization of the training process is shown in Figure 1a. The number of input and output neuron nodes was determined by 39 engineering parameters (Appendix A.

Table A1. The 39 engineering parameters.

No.	Name	No.	Name	No.	Name
1	Weight of moving object	14	Strength	27	Reliability
2	Weight of stationary object	15	Durability of moving object	28	Measurement accuracy
3	Length of moving object	16	Durability of non-moving object	29	Manufacturing precision
4	Length of stationary object	17	Temperature	30	Object-affected harmful
5	Area of moving object	18	Illumination intensity	31	Harmful side effect
6	Area of stationary object	19	Use of energy by moving object	32	Ease of manufacture
7	Volume of moving object	20	Use of energy by stationary object	33	Ease of operation
8	Volume of stationary object	21	Power	34	Ease of repair
9	Speed	22	Loss of Energy	35	Adaptability of versatility
10	Force	23	Loss of substance	36	Complexity of device
11	Stress of pressure	24	Loss of Information	37	Complexity of control
12	Shape	25	Loss of Time	38	Level of automation
13	Stability of the object	26	Quantity of substance	39	Productivity

) and nine technological evolution laws (Appendix A.

Table A2. The nine technological evolution laws.

No.	Name
1	Law of Increasing degree of ideality
2	Law of non-uniform evolution of sub-systems
3	Law of Increasing dynamism
4	Law of transition to a higher-level system
5	Law of transition to micro-level
6	Law of completeness
7	Law of shortening of energy flow path
8	Law of increasing controllability
9	Law of harmonization

). Part of the training cases are shown in

Table 1. Part of the training cases.

No.	Case	Different Stages	EP and Order	TEL
1	Air conditioner	Fixed frequency air conditioner → Inverter air conditioner → Central air conditioning → Energy storage	27.1 26.2 22.3	8
2	Automotive control system	Engine control → Manual transmission → Automatic gear control → infinitely variable speed	31.1 33.2 35.3	3
3	Glasses	Colorless glasses → Sunglasses → Discolored glasses	18.1 35.2	5
4	Airbag	compressed gas → chemical reaction → Coanda effect	15.1 31.2	7
5	Peeler	Blade → Stir peeling → Electric peeling → Adaptive peeling	33.1 38.2 35.3	6

EP denotes the engineering parameter. TEL denotes the technological evolution law.

. 20 cases were randomly selected from 100 cases for testing the accuracy of the training model. The test results are shown in Figure 1b. The prediction accuracy of the model can reach 85%. The accuracy can be further improved by increasing in the training set.

The process of obtaining the technological evolution law for predicting potential features according to the BPNN is shown in Figure 2. The product is divided into four subsystems according to the energy, transmission, actuation and control devices. Based on the development vision, the company selects from the above subsystems for improvement as the target objects. The engineering parameters corresponding to different phases of the target object development and their order of appearance are extracted. The best technological evolution law is predicted by the trained BPNN model, engineering parameters and their order for potential features. The acquisition of potentially new technologies and solutions are discussed in following sections.

2.2. Acquiring Potentially New Technologies Using the IPC Knowledge

2.2.1. IPC Design Knowledge

Design knowledge in the IPC is embedded in the subclass, main group and subgroup, containing functions and effects. The function is an abstract description of the component action. It is usually expressed as a form of verb + noun [72,73]. However, some functions are not expressed in a standard form and often appear as the functional verb in the IPC. Therefore, functions in the IPC are uniformly expressed in the form of the functional verb. The set of functional verbs that appear in each subclass and following groups is defined as IV. V_i are the functional verbs in a set of IVs.

The relationship between V_i co-occurs at each hierarchy of the subclass and subsequent group, as shown in Figure 3a. There are three relationships between functional verbs. Similarity, the function represented by functional verbs that has a similar relationship with some overlap between functions. V₁ ∩ V₂ ≠ ∅. For example, B26F: perforate/punch. Auxiliary, the function indicated by the functional verbs that are auxiliary to each other or to the preceding or following processes. V₃ ∪ V₄. For example, F24F13/32: humidify/ventilate. Opposite, the function indicated by the functional verbs that are opposite. V_i−1 ∩ V_i = ∅. For example, H03K5/04, increase/decrease.

The effect is a process description of the functional transformation of the system. It consists of laws, axioms, and principles from physics, chemistry, geometry and other subjects [74]. The effect is usually not given directly in the IPC, but is described using a certain format of phrases [53]. For example, ‘noun + verb’, ‘noun + action’. In a set IE of effects under each class, E_j are the effects in a set of IEs.

The relationship between E_j that appears in each subclass is shown in Figure 3b. As with functional verbs, there are three relationships between effects.

2.2.2. Relevancy and Universality

The following terms are defined for the method description. The verb corresponding to the current function is called the current verb. Verbs that co-occur with the current verb in each subclass and following groups are called associative verbs. The effect of achieving the current function is called the current effect. The effects related to the current verb and associative verb or that co-occur with the current effect under a subclass are called associative effects. Since the number of effects is small, texts of the subclass, main group and subgroup are combined to extract co-occurring effects. An example of ‘liquid cool’ is shown in Figure 4 to describe the uses of the terms in the IPC.

The subclass, main group and subgroup in the IPC are divided by similar structures of patents. A large number of the co-occurrence of functional verbs at each level in each subclass and following group means that the function similarity or number of combinations in the previous technology is high. The technology is more likely to be replaced or combined. The relevancy of the functional verbs is as follows.

$$A_v={\displaystyle \sum N_{scvi}}+{\displaystyle \sum N_{mgvj}}+{\displaystyle \sum N_{sgvk}}$$

(1)

where $A_v$ is the relevancy of the functional verbs; $N_{scvi}$, $N_{mgvj}$ and $N_{sgvk}$ denote the number of the co-occurrences of the current and associative verbs in the subclass, main group and subgroups, respectively.

A process of the NLP-based associative verb extraction and search is shown in Figure 5.

It consists of the following steps.

• Section and class information of the IPC downloaded from the National Intellectual Property Office [75] is removed.
• In the Spyder-integrated development environment, LTP is used for word segmentation and part-of-speech tagging. After removing stopwords and repetitive words, functional verbs are extracted as part-of-speech tagging.
• Functional verb relationships extracted from each subclass, main group and subgroup are established. The same order number (e.g. 1, 2, 3, 4…) in the build relationship of Figure 5 represents the same category, and V_i represents the different functional verbs extracted from that category. The functional verb relationship is established using the Bibexcel software.
• Fully connected networks are built with verbs in each category as nodes and co-occurrence relationships as edges. The Networkx module is used to build fully connected networks in Spyder.
• Based on the edge information in the fully connected networks and Equation (1), the current functional verb is input into the search program to extract associative verbs and relevancy. The search process incorporates the input of the minimum relevancy value to filter the associative verbs with smaller relevancy. The minimum relevancy value is determined based on the initial search results. The Matplotlib module is used to implement the visualization in Spyder.

A large number of effect co-occurrences in the subclass mean a high similarity or number of combinations in previous technology. It is more likely to be replaced or combined. The effect relevancy is decided by Equation (2).

$$A_e={\displaystyle \sum A_{ei}}$$

(2)

where $A_e$ represents the relevancy of the effect; $A_{ei}$ represents the relevancy of the effect within each subclass.

The process of NLP-based associative effect and effect relevancy uses the same steps of the functional verb extraction method. Only the text processing and effect extraction rules are different, as follows.

• The information of the section and class in the IPC is removed. The text information of the subclass, main group and subgroup is combined.
• In the Spyder-integrated development environment, the word segmentation and part-of-speech tagging in LTP and effect extraction rules are used to extract effects and filtering effects by loading stopword dictionaries and removing repetitive effects. The effect extraction rules are shown in

  Table 2. Effect extraction rules.

  Rule	Composition Type	Extraction Result of Effect
  1	a noun and a verb. n + v	n + v
  2	multiple nouns and a verb, the last noun is connected by ‘or’. n1, n2, …, ni−1 or ni + v	n1 + v; n2 + v; n3 + v; n4 + v; n5 + v
  3	multiple nouns and a verb, the last noun is connected by ‘and’. n1, n2, …, ni−1 and ni + v	n1 + v; n2 + v; n3 + v; n4 + v; n5 + v
  4	a noun and multiple verbs, the last verb is connected by ‘or’. n + v1, v2, …, vi−1 or vi	n + v1; n + v2; …; n + vi−1; n + vi
  5	a noun and multiple verbs, the last verb is connected by ‘and’. n + v1, v2, …, vi−1 and vi	n + v1; n + v2; …; n + vi−1; n + vi
  6	words contain ‘effect’, or ‘action’., e.g., ‘hall effect’. (Note, LTP divides it into one noun. Chinese to English translation will show two words.)	words containing effect or action
  7	a noun and ‘effect’. n + effect	n + effect
  8	a verb and ‘effect’. v + effect	v + effect
  9	a noun and ‘action’. n + action	n + action
  10	a verb and ‘action’. v + action	v + action

  .

Since a function can be represented in the form of verb + noun, some functions cannot be directly linked to associative effects in the IPC. Some effects related to associative verbs cannot co-occur with the current effect in a subclass. For these cases, the relevancy may not be decided by Equation (2). Therefore, the universality [53] is used to select the effect, as shown in Equation (3). The effect with a high universality indicates better suitability in the applied field, which is better for the reapplication of the technology.

$$U_e={\displaystyle \sum U_{ei}}$$

(3)

where $U_e$ is the universality of an effect; $U_{ei}$ denotes the number of effect occurrences in each subclass.

Since the relevancy can represent the relevance of technologies, the universality can only indicate suitability in existing domains. Therefore, when the relevancy is available, it is preferred for the selection of effects.

The process of NLP-based associative effect and universality is as follows:

• Texts of subclass, main group and subgroup are combined. Effects under each subclass are extracted in the Spyder-integrated development environment. After filtering and removing repetitive effects, they are used as a search database for universality.
• Functional verbs and the maximum number of associative effects to be displayed in the word cloud are input into the universality search program. According to Equation (3), the larger the universality, the larger the font size of the associative effect in the word cloud. The Matplotlib and word cloud modules are used to implement the visualization in Spyder.

2.2.3. Acquisition of Potentially New Technology

The process of acquiring potentially new technology is shown in Figure 6.

It consists of following steps.

• Extracting current verbs of the design target. According to the function of the design target and corresponding functional verbs in the IPC category where it belongs, the current verb is determined.
• Searching and identifying associative verbs. The functional verb search program is run to obtain the associated verbs. The associative verbs that meet the working conditions are determined according to their potential features and relevancy. Subsequently, Step 3 can be omitted if the associative verbs already satisfy the search for a potential solution.
• Searching and identifying associative effects. Current verb, current effect, or identified associative verbs are input into the effect search program. The associative effects that meet the working conditions are determined according to their potential features, effect relevancy or universality.

2.3. Search and Application of the Best Potential Solution

The potentially new technologies are identified through the functional verbs or effects in the IPC, but the solutions to realize these potentially new technologies are not identified. Therefore, the patent database is used to search for potential solutions to analogize.

2.3.1. Selection of the Best Company

Different patents may use different language to describe functions and features of potentially new technologies. A semantic expansion [76] of functions and potential features is required to increase the comprehensiveness of the patent search. The keyword assistant in the PatSnap patent database [77] is used to semantically expand verb and noun functions, and potential features. The search for potential solutions is then performed with the expanded keywords.

The list of patents after searching by keywords will show multiple companies related to the potentially new technology. The more patents that a company applies for the same technology indicates a higher maturity of that technology, but it does not mean that the company has the most suitable solution for the potentially new technology of the target product. The greater the overlap of patents applied for under the same IPC between companies indicates a closer technological distance, which means that the technology similarity is high and the technology is easily absorbed. A large technological distance indicates a high technological novelty.

The relationship between innovation outputs and a company’s technological distance have an inverted U-shape. This result is derived from the empirical analysis of a binomial regression model where the innovation output peaks at a technological distance of 0.4 [78]. However, in the actual process, the technological distance of the best company is difficult to take the value of 0.4. In particular, when a potentially technology that is predicted by the technological evolution law have a large gap with the current technology, there will be a high technological distance between the companies. Therefore, the company with the smallest absolute value with 0.4 is selected as the best company for the potential solution in this research.

$$D\mathrm{istance}=1-\frac{{\displaystyle {\sum }_{k=1}^{131}{f}_{ik}{f}_{jk}}}{\sqrt{{\displaystyle {\sum }_{k=1}^{131}{f}_{ik}^2{\displaystyle {\sum }_{k=1}^{131}{f}_{jk}^2}}}}$$

(4)

where Distance denotes the company’s technological distance; $f_{ik}$ represents the number of patents of company i in k field; and $f_{jk}$ represents the number of patents of company j in k field. According to the latest IPC there are 131 class.

2.3.2. Solution Mapping

The mapping of potential solutions from the best company to a target product is based on the improvement or overlay of the current technology by new technology. The functional model in TRIZ [79] is first used to establish the relationship between the components of the current product technological system (TS). The mapping relationship between the TS, potential technology (PT) and potential solution (PS) is then established as shown in Figure 7, where C represents the components of each system, and arrows indicate the interaction between the components.

2.3.3. Evaluation of Design Solutions

The best company may have multiple technological solutions for the same technology. After the aggregation of the solutions, a summary solution with the highest overall index should be selected as the final innovative design solution. The idealization level of the product is then introduced [80]. Ideality is calculated based on the average value of each evaluation index. The evaluation interval is set as 1–5 for m evaluators and n design solutions. The calculation process is as follows.

$${\overline{x}}_{kj}=\frac{{\displaystyle \sum x_{kij}}}{m}$$

(5)

$$I_j=\frac{{\overline{x}}_{Uj}}{{\overline{x}}_{Hj}+{\overline{x}}_{Cj}}$$

(6)

where $x_{kij}$ denotes the evaluation of the kth indicator of the jth solution by the ith evaluator k is U, H, and C. U denotes the useful feature; H denotes the harmful factor; and C denotes the cost); ${\overline{x}}_{kj}$ denotes the average kth indicator value of the jth solution; $I_j$ denotes the ideality of the jth solution; ${\overline{x}}_{Uj}$ is the average value of the useful feature evaluation index brought by the function of the jth solution; ${\overline{x}}_{Hj}$ is the average value of the harmful factors evaluation index brought by the function of the jth solution; and ${\overline{x}}_{Cj}$ is the average value of the cost evaluation index of the jth solution.

2.4. Framework

The proposed method was used to stimulate innovative product design based on the BPNN, IPC knowledge and company’s technical distance to acquire technological evolution potential. It consists of the following three steps, as shown in Figure 8.

• A target product is decomposed into subsystems according to energy, transmission, actuation, and control devices. Multiple stages of the subsystem update process are determined through patent analysis of the target product within the company. Engineering parameters are extracted for each stage of improvement, which is input into the BPNN model in order to find the technological evolution law for determining the potential features of the product.
• Functional verbs of the design target function in the IPC are identified. Functional verbs are the input of the search program to search for associative verbs. The associative verbs used for the potential state are determined based on the potential features, relevancy and working conditions. If the associative verbs are not sufficient to search for technical solutions, the effect search program is used to search and identify associative effects for accessing potentially new technologies.
• The keyword assistant in the PatSnap database is used to semantically expand functions of potentially new technology and potential features for determining keywords. Semantically expanded keywords of the potential technologies are used to search for technical solutions. A company’s technological distance is used to select analogous sources of the best company technology solution. The new design of the product is carried out by solution mapping. If multiple innovative solutions are formed, the final design is selected through the evaluation.

3. Case Study

There are various models of blade cutting machines on the market using a similar working principle. A new generation of the steel pipe cutters has been proposed for Suzhou Baoshide Electric Tools Co., Ltd. (Suzhou, China) in a case study to illustrate and verify the proposed method.

3.1. Analyzing the Potential Features of the Product

Sixty-seven invention patents of company cutting machines are searched from the patent database. The actuation device of the cutting machines was selected as an example to illustrate and verify the proposed method. The components associated with the actuation device are the clamp or workbench for fixing the steel pipe and blade. Four representative patents were selected to analyze the improvements of each product generation of the actuation device, as shown in Figure 9. The main clamp was continuously improved in the actuation device. From Figure 9a to b, rollers are installed under the clamp in order to improve moving difficulties, the corresponding engineering parameter is 33, ease of operability. From Figure 9b to c, the position of the clamp relative to the blade can be moved for different sizes of steel pipes, the corresponding engineering parameter is 35, adaptability or versatility. From Figure 9c to d, the clamp guide slot is changed from line contact to surface contact to increase accuracy, the corresponding engineering parameter is 29, manufacturing precision. The above engineering parameters were sequentially input into the BPNN model, as shown in Figure 10. The technological evolution law was obtained. Therefore, the potential feature of the actuation device should evolve to a higher-level system and continue to increase the diversity of the components.

3.2. Acquiring Potentially New Technologies Using IPC Knowledge

The function performed by the clamp is ‘fix steel pipe’. The current verb of the function in the IPC is ‘fix’. The ‘fix’ and ‘minimum relevancy’ are input into the functional verb search program to obtain the associated verbs, as shown in Figure 11. The number on the line represents the relevancy. The input functional verb and associative verbs have been processed by translating them from Chinese to English.

According to the search results, the top five associative verbs were rotate, move, adsorb, suspend, cool and drive. The associated verb, rotate, allows the steel pipe to change from a fixed state to a rotating state, evolving to a higher-level system. The associative verb, move, can be used for the moving clamp, which is already owned by the company and is no longer considered. The associative verb, adsorb, can be used to adsorb irritant gases generated by cutting the steel pipe, increasing the diversity of the actuation device. The associative verb, suspended, has no associative function to match it in the cutting state. The associative verb, cool, can be used as a subsequent process when the steel pipe has been cut. The associated verb, drive, allows the clamp to be changed from manual to electric driving, but this belongs to the 6th technological evolution law, which is not consistent with the company’s product development direction.

Since this case is only used to illustrate and verify the feasibility of this method, only two associative verbs with the most relevancy to the working conditions, rotate and adsorb, were chosen for subsequent processes.

The associative verb, rotate, already indicates the potential state of clamping the steel pipe in a way that does not require to search for the effects. It is enough to find the technical solution to achieve this potential state. For the associative verb, adsorb, there are different effects to achieve it. Further analysis of the effect is needed. Since the relation between ‘adsorb’ and ‘fix’ is auxiliary, it is a newly added function, effects related to adsorb could not be found in the current function. The universality is used to search the associative effects. The ‘adsorb’ and ‘maximum number of effects’ are input into the universality search program to obtain the associative effects, as shown in Figure 12. All associative effects contain the word ‘adsorb’ in Chinese. The representation of the input functional verb and associative effects are translated from Chinese into English.

According to the word cloud, it can be seen that effects with the most universality are ‘solid adsorb’ and ‘vacuum adsorb’. Since ‘solid adsorb’ (e.g., activated carbon) is not suitable for fast-absorbing scenarios, ‘vacuum adsorb’ is chosen for the potentially new technology.

3.3. Search and Application of the Potential Solutions

3.3.1. Selection of the Best Company

• Best Company of Rotation

Although the steel pipe changes from a fixed state to a rotating state, the function is still to ‘fix steel pipe’. The keyword assistant of the PatSnap patent database is used to assist the semantic expansion of the verb ‘fix’, the noun ‘steel pipe’, and the potential feature ‘rotate’. The expanded keywords are used to construct a search formula for the patent search. The top 15 companies are selected as candidates. The number of relevant patents and all invention patents of each company are listed in Appendix A,

Table A3. Patents of the 15 companies (rotation).

Code Name	Company	Number of Relevant Patent	Total Number of Invention Patent
A1	State Grid (Beijing, China)	10	37,971
B1	China Heavy Machinery Research Institute (Shaanxi, China)	6	558
C1	Nippon Steel Co. (Shanghai, China)	6	16,171
D1	Wechtolli Co. (USA)	4	1001
E1	Changzhou Tenglong Auto Parts Co. (Changzhou, China)	4	9
F1	China Twenty Metallurgical Group Co. (Shanghai, China)	4	554
G1	China Nuclear Technology Development Co., Ltd. (Tianjin, China)	4	74
H1	Baowu Special Metallurgy Hang Research Technology Co. (Chongqing, China)	4	109
I1	China National Petroleum Co. (Beijing, China)	4	5508
J1	Shanghai Wuye Group Co., Ltd. (Shanghai, China)	3	499
K1	Shandong Water Diversion Engineering Technology Research Center (Shandong, China)	3	8
L1	Jiangsu Zhonghai Heavy Machine Tool Co., Ltd. (Jiangsu, China)	3	25
M1	Mit-subishi Heavy r Industries Ltd. (Beijing, China)	3	12,661
N1	MolexInc (USA)	3	3760
O1	Wanshitai Metal Industry (Kunshan) Co., Ltd.(Jiangsu, China)	3	14

. The patent number of each company in each class is exported by the analysis module of PatSnap. According to Equation (4), the MATLAB software is used to calculate the technological distance between 15 companies and Suzhou Baoshide Electric Tools Co., Ltd. (Suzhou, China), as shown in Figure 13a. The company with the best technological distance was G1, China Nuclear Technology Development Co., Ltd. (Tianjin, China).

2.

Best Company of Vacuum Absorption

The effect performs the function of adsorbing gas. The keyword assistant is used to assist the semantic expansion of the function and potential feature. The search formula is constructed for the patent search. The top 15 companies are selected as candidates. The number of relevant patents and all invention patents of each company are listed in Appendix A,

Table A4. Patents of the 15 companies (vacuum absorption).

Code Name	Company	Number of Relevant Patent	Total Number of Invention Patent
A2	Royal Philips Electronics Co., Ltd. (Netherland)	99	142,645
B2	Reesmead Co., Ltd.(Australia)	72	1585
C2	Engineering Inspiration Company (Italiana)	51	1455
D2	Haier Zhijia Co., Ltd. (Qingdao, China)	35	5985
E2	China Petroleum and Chemical Co. (Zhejiang, China)	34	31,015
F2	Zhuhai Geli Electric Co., Ltd. (Guangdong, China)	33	13,803
G2	Mei Group Co., Ltd. (Guangdong, China)	32	9469
H2	Fisher & Paykel Healthcare Co., Ltd. (Guangdong, China)	30	1386
I2	Honda Technical Research Industrial Co. (Shanghai, China)	25	43,688
J2	Guangdong Meizhi Refrigeration Equipment Co., Ltd. (Guangdong, China)	25	602
K2	Beijing North Huachuang Microelectronic Equipment Co., Ltd. (Beijing, China)	24	2136
L2	General Electric Company (USA)	23	79,295
M2	Deerge Manufacturing Co., Ltd. (Germany)	23	3580
N2	Pacific Water Mud Co. (Japan)	23	422
O2	Tokyo Electron Ltd. (Japan)	20	9278

. The technological distance between the 15 companies and Suzhou Baoshide Electric Tools Co., Ltd. (Suzhou, China) is calculated, as shown in Figure 13b. The company with the best technological distance was L2, General Electric Company (Boston, MA, USA).

3.3.2. Potential Solution Mapping

The China Nuclear Technology Development Co., Ltd. (Tianjin, China) possess four relevant patents. However, after analysis, only two were actually available, and both are around the same technology for patent layout. The technical implementation is shown in Figure 14a. The General Electric Company (Boston, MA, USA) possess twenty-three relevant patents. After analysis, only two patents are actually available, as shown in Figure 14b,c.

The mapping relationship between TS, PT and PS is established, as shown in Figure 15.

The technical solution of a steel pipe rotation according to PS₁ is shown in Figure 16a. The chuck and rotatory mold are analogous to the fixing device of uncut steel pipes and the collection device of cut steel pipes. Since this patent requires different types of rotatory molds for different sizes of steel pipes, it makes the operation process tedious. Therefore, the fixed-size rotatory mold is replaced by a variable size collection device. The guide rails and skid platform are analogous to the longitudinal and lateral movement guide rails of the steel pipe collection device. The motor and main shaft box are analogous to the drive device and transmission.

The adsorbing potential solution is formed according to PS₂₁, as shown in Figure 16b. The pressure machine creates a negative pressure environment for adsorbing irritant gas. The water in the humidifier is replaced with a chemical that produces a neutralizing reaction with the irritant gas. The filtration membrane is used for the collection of iron filings to avoid damaging the machine.

The adsorbing potential solution is formed according to PS₂₂, as shown in Figure 16c. A turbo is analogous to a negative pressure machine for adsorbing irritant gas. The turbine engine is analogous to a combustion chamber for irritant gas elimination by burning it. A filter is used for the treatment of exhaust gases after fuel combustion.

3.3.3. Evaluation of Design Solutions

Technological solutions for each of the above potential state are combined to form two summary results, as shown in Figure 17.

The emergence of a new generation of products proposes a corresponding performance improvement. The newly added function, adsorbing the irritant gas, significantly improved the operating environment. In a fixed state, cutting the steel pipe is only subjected to a force in one direction. The deformation will gradually increase with the increase in cutting depth. The steel pipe is evenly stressed around in the rotating state. The deformation can be properly improved. At the same time, the cutting depth of the blade only needs to exceed the thickness of the steel pipe; thus, the cutting diameter range has been increased. Therefore, the potential technology not only changes the actuation device, but also improves the performance of the cutting machine.

Five evaluators, M, (two users, three engineers) were invited to evaluate the two summary solutions. Where, S₁ denotes the summary solution 1, S₂ denotes the summary solution 2. In addition, the ideality of the two new summary solutions was also evaluated and compared to the original design S₀ (Fourth generation in Figure 9). The evaluation of the solutions is shown in

Table 3. Solution evaluation.

Members	U			H			C
	S0	S1	S2	S0	S1	S2	S0	S1	S2
M1	3	5	5	4	1	2	1	3	4
M2	2	4	4	4	2	2	2	5	5
M3	2	5	5	5	1	2	1	3	4
M4	1	3	3	4	1	1	2	4	5
M5	3	5	5	3	1	1	1	4	5

.

The ideality of each solution is calculated according to Equations (5) and (6), $I_{S_0}=0.407$, $I_{S_1}=0.880$, $I_{S_2}=0.710$. It can be found that both new design solutions are better than the original product. The evaluation of S₁ is better than that of S₂. Therefore, S₁ is selected as the final innovative design.

4. Discussion

A new BPNN model was trained for the selection of the technological evolutionary law of replacement and update. The prediction accuracy reached up to 85%. Based on the test result analysis, the prediction errors were due to the fact that the randomly selected training samples did not contain similar combinations of engineering parameters. However, based on the analysis of its correctly predicted test set, its prediction results were acceptable. At the same time, for the BPNN training cases, the technological evolution laws corresponding to each case were analyzed jointly by domain experts, which reduces the risk of selecting the technological evolution laws due to subjective errors compared to designers who cannot fully grasp the technological evolution laws. Therefore, this model is feasible for the selection of technological evolution laws.

Function and effect knowledge in the IPC were extracted under the Spyder-integrated development environment based on regular expressions, which was used to build the technology potential search database. The best analogous source in the patent database was searched by the company technology distance. This process was verified using a steel pipe-cutting machine as a case study. The new design solution was compared with the original solution by an ideal solution, the feasibility of the method was verified.

There are some limitations of the proposed method as follows. (1) Due to the limited training samples of the BPNN, there were errors in the prediction results of some combinations of the engineering parameters. (2) The accuracy of the extracted functional verbs or effects can be affected due to the accuracy of the LTP word segmentation. (3) Different types of innovative design solutions may be generated by the technological evolution in different product life cycle stages. The proposed method does not consider the life cycle of the product. (4) The process of using the BPNN to predict the technological evolution law requires at least two engineering parameters as an input. The BPNN method proposed in this study is not applicable to brand-new-listed products as it cannot extract the required number of engineering parameters.

Future work will expand the training set to improve the accuracy of the prediction results. With the improvement of the LTP word segmentation, the functional verbs and effects in the IPC will be re-extracted to improve their accuracy. Data mining will be used to predict and manage the product life cycles in analyzing the type of innovation design potentials by the technological evolution law. For brand-new-listed products, the corresponding engineering parameters will be extracted from the product functional requirements based on NLP technology for predicting the technological evolution law. At the same time, the development of a complete software tool will also be useful in future work.

5. Conclusions

Contributions of this research are as follows. Theoretical aspects: (1) BPNN can be used to select the best technological evolution law corresponding to product replacement and update. It reduces the subjectivity of designers and extends the application of the BPNN to technological evolution laws. (2) Functional verbs and effects in the IPC can be extracted and searched by using NLP and improved rules, providing feasibility for its knowledge utilization. At the same time, the potential features provides a basis for the preliminary filtering of irrelevant functional verbs or effects in the IPC. (3) IPC knowledge and a company’s technological distance can be used for the acquisition of potential states, which can improve the existing research to access potentially new technologies and solutions. The proposed method is not only conducive to designers to generate new concepts and overcoming the reliance on design experience, but also to help them to apply cross-domain potential solutions in patent databases with huge analogous sources. Practical aspects: (1) Challenges of the product innovation process are idea generation and cross-domain knowledge application. This research proposes a systematic approach from concept generation to cross-domain solution finding, which addresses the problem of cross-domain applications of knowledge in industry. (2) For the technology innovation management, this research provides new methodological insights into the application of cross-domain technology, which provides a new way of thinking about its direction in guiding the development of industrial products. (3) The intellectual property department and design department of a company are usually independent. Since design departments are biased towards product design, they lack effective access to knowledge retrieval. The intellectual property department favors technology protection and blocking. This method can be applied by the intellectual property department as well as serving the design department to improve the efficiency of product development.

In order to improve the process of technological evolution to stimulate innovative product design, this paper combines the technological evolution laws with the BPNN, IPC knowledge and a company’s technological distance. According to improvement measures at different stages of product development, corresponding parameters and their order of appearance were extracted. A BPNN model was built to select the best technological evolution law based on engineering parameters and their order. The knowledge of the functions and effects contained in the IPC was introduced in this process. The functional verbs and effects co-occurring in the IPC were used to stimulate the acquisition of potentially new technologies. At the same time, the company’s technology distance was proposed to measure analogous sources of potential solutions in a patent database with a large amount of design information. The ideality was used for the selection of the final innovative solution. A steel pipe-cutting machine was developed to verify the feasibility of the proposed method. In summary, for effective access to technological evolution for stimulating innovative product design, this paper proposes an integrated method and a rational technological path.