A Knowledge Push Approach to Support the Green Concept Design of Products

Abstract

With the development of the manufacturing industry, environmental problems have become more and more serious. At present, green design is playing an increasingly important role in product design and manufacturing. How to combine traditional conceptual design methods with green designs to realize green conceptual design is of great practical significance. In order to realize the green concept design of a product and to allow the conceptual design scheme to meet the functional requirements and green performance requirements of the product at the same time, this study introduces the knowledge push process and the green filter process into product concept design. The knowledge of invention principles is retrieved by establishing the knowledge space of invention principles, by calculating the word segmentation based on the CRF (conditional random field), and calculating the word similarity using a thesaurus. This is performed in order to realize the transformation from product function requirements to invention principle knowledge. Then, by studying the relationship among the environmental efficiency factors, green design attributes, engineering technical parameters, and invention principles, the correlation mapping table for green design attributes and invention principles is established. Thus, the green filter is applied to the pushed knowledge according to the green performance requirements, so that the final push results after filtering can meet the functional and green performance requirements of the product. In the last section of this paper, a green concept design for cutting the fluid supply system in an enterprise is presented as an example to verify and analyse the proposed method.

1. Introduction

With the rapid development of the manufacturing industry, environmental problems have become increasingly prominent. Currently, wastewater, exhaust, and waste generated by the manufacturing industry account for more than 70 per cent of global pollution emissions [1], and it is increasingly important and urgent to reduce these emissions to achieve global sustainability. In order to solve the environmental problems brought about by the rapid development of the manufacturing industry, many countries have put forward higher requirements for enterprises with respect to green design and manufacturing. At the same time, research related to green concepts in computer science, engineering, environmental sciences, and industry has been carried out and developed at the international level over the last few years [2]. Karabetian et al. [3] proposed a Dimensioning Workbench service to reduce carbon emissions. Saxena et al. [4] presented a review of green computing. In addition, some studies have shown that the design phase of a product is widely regarded as the key stage in the product life cycle to achieve the green attributes of a product [5]; about 70–85% of a product’s cost is determined during the conceptual design phase, but its cost is only about 5% of the product development resources [6]. Obviously, whether or not green design concepts are included in a product’s conceptual design can have a significant impact on the subsequent design process and product manufacturing, performance, cost, energy consumption, and pollution, and determines whether or not the product can ultimately achieve full-life-cycle sustainability, as well as the degree to which green design is present in the final product.

However, the conceptual design of products requires a great deal of knowledge, which is often interdisciplinary, multi-scoped, and complex, and the designer often does not have complete knowledge of a wide range of areas. At the same time, most designers in this stage are still stuck in traditional conceptual design approaches, which lack the concept of green design, and so they can often only complete the basic structure of the product, but cannot effectively combine the product attributes with green attributes, so it is difficult to meet the user’s green needs for the product. Traditional product design theories, such as the theory of inventive problem solving (TRIZ) proposed by Altshuller et al., generalises the problem solving patterns for millions of invention patents. It provides the paradigms, methodological systems, and tools for analysing and solving problems, and enables the knowledge of invention principles to be obtained through conflict matrices or separation principles. Although the theory enables the acquisition of the knowledge of invention principles, the processes from knowing the parameters to knowledge and from having knowledge to the solution are still complicated by the need for designers to find the appropriate parameters themselves, and by the abstract nature of the knowledge of invention principles. And the theory does not include the green design concepts with which to achieve the green attributes of a product. Similarly, for the conceptual design process of a product, Gero et al. [7] first proposed the FBS (function–behaviour–structure, FBS) conceptual design model, which presents a design process that links function, behaviour, and structure together and treats them as different stages of the design process; the later-contextualised FBS model extends the original 8 mapping relationships to 20. Based on this classic model, related studies have developed the FBS Path model [8], the ESBF model [9], the FSMEE model [10], the FPBS model [11], and the RFBS model [12].In addition to these, Chen et al. [13] proposed a model for a requirements–function–principle-system conceptual design process, which includes clarification, synthesis, implementation, analysis, and prediction phases, through a normative definition of the relevant design concepts. Li et al. [14] proposed an integrated, conceptual design process model containing five phases and four mappings, and used a mathematical language to describe the integration logic of the process model. Camelo et al. [15] proposed an interactive-product conceptual design process model. This model can extend the search space for design solutions by expanding the relationships between the design elements. Agung et al. [2] proposed a classification for the convergence axis, which includes publications in green computing, to categorize the body of knowledge produced by forty-one years of academic publications in terms of their knowledge contributions: computer science, environmental management, mobile computing, energy, and sustainability, abbreviated as CEMES study themes.

The product concept design models mentioned above, as traditional design methods, are similar to the TRIZ theory in that they are mostly centred on normative research and lack the appropriate green concepts to complement the design of models when faced with green needs.

Accordingly, a large number of improved conceptual design models have been proposed based on the above-mentioned need for green design and the improvement of traditional theoretical models. Fu Yan et al. [16] proposed a function–structure–material-process green design model to support the generation of green whole-life design solutions for products. Umeda et al. [17] proposed a scalable product design approach based on the FBS model to address environmental issues, especially resource depletion. Lei Zhang et al. [18] established a multi-level iterative-process model for function–behaviour–structure green design units, and proposed a knowledge-based reuse method for green product concept design to solve the problem of long implementation cycles and low efficiency in green design. Jeong et al. [19] proposed the functional–behavioural–structural–environmental effects (FBSe) expression model for describing the impact of products on the environment, and completed a case-based reasoning for product eco-design. The above studies, based on improved FBS models, can support the green design modelling of products, but they either focus on specific materials and processes in product design or provide feedback to correct the design after the overall structure of the product has been designed; none of them consider green design at the principle level of product design. At the same time, they all lack a matching knowledge push, and their product green design processes need to be further improved and optimised. The knowledge recommendation aspect of product design has also been studied in depth by many scholars. Sarica et al. [20] proposed the use of technical semantic networks to address design knowledge representation. Murphy et al. [21] proposed a functional vector approach to encode design knowledge by representing design documents as vectors based on the word frequencies of functional verbs in a functional base model. Sanaei et al. [22] proposed a method for extracting relational structures from a text through text mining and RNN-based models for retrieval mapping in the process of knowledge pushing. Jin et al. [23] proposed a knowledge support approach to put users in the designer’s shoes and generate user-centred design inspiration. And, in addition, they developed four computational models for analogies that capture the relational structure of the text. This includes the spatial representation of semantics, multi-level deep neural reasoning, a graph matching-based model, and a transformation-based model. The models were then combined together into an ensemble model to achieve an acceptable level of analogical accuracy for the end user. All of the above studies can support the whole knowledge push process, but there are fewer studies that match and link the knowledge push process to the conceptual design process, and there is a near total lack of studies that consider the integration of green design and knowledge push.

As can be seen from the above, a large number of scholars have conducted research on knowledge push and traditional conceptual design methods, which have to some extent achieved a combination of conceptual design and green design, as well as matching knowledge push systems. However, due to the more ambiguous, diverse, and dynamic nature of user green needs in practice, there is an even greater need for principle-level improvements to the model of green product concept design in production practice, the integration of traditional design theory with green design concepts, and the combination of concept design theory with knowledge push theory. To address these issues, this paper adopts the knowledge of invention principles based on the traditional FPBS model, thereby guiding designers in the discovery of conceptual solutions from the use of engineering design laws. Through the established design model, the knowledge of inventive principles is pushed and green-filtered, and eventually the knowledge pushed is in the form of patented knowledge, so that the final solution results can meet the product’s functional requirements, structural requirements, and green requirements at the same time.

2. Green-Concept-Design Knowledge Push Model for Product Development

In the knowledge transfer process in product concept design, in addition to the theoretical model used to determine the design process, it is also necessary to consider the specific methods used to complete the implementation of each step of the process. Therefore, the FBS mapping model is used here as the classic model structure, on the basis of which the variable behaviour is introduced to obtain a more complex and improved model-mapping structure of function–principle–behaviour–structure. The knowledge push model studied in this paper is divided into an FPBS process layer, a method layer, and an output layer, each of which is interlinked. The FPBS process layer includes the various stages from the requirements to concept solution generation, the method layer mainly serves to support the implementation of the various stages of the process layer, and the output layer gives the outputs of the various stages of the process layer, as shown in Figure 1.

The knowledge pushing model combines the theoretical design process and the applied design knowledge, pushing the design knowledge of invention principles together with the FPBS theoretical design process and, at the same time, green-filtering the knowledge of invention principles according to green performance requirements, realising a knowledge pushing model for green performance requirements. Specifically, it is divided into the construction of the invention principle knowledge space and the application of the FPBS design process. The construction of the invention principle knowledge space mainly occurs through the svop functional semantic decomposition of the invention principle to establish the knowledge space, while also screening and classifying its green design attributes. The theoretical design process uses the FPBS conceptual design process, using the CRF algorithm to analyse the functional requirements using word separation, and then uses the word similarity algorithm to achieve knowledge retrieval mapping, to assist in the knowledge push of the invention principle, the knowledge support for the conceptual design, to facilitate the knowledge reuse of the designer, and to achieve the conceptual design solution generation.

2.1. Principles of Invention in Knowledge Space Construction

The application of the design knowledge space established in this paper is mainly the knowledge of invention principles, which are connected and extended with nodes to realise the construction of a wholly designed knowledge space, which is finally displayed in the form of a diagram. The process of applying invention principles is the process of realising functions. It was found by [24] that the invention principle can be expressed semantically as v (functional action) + o (functional receptor) + p (attribute parameter). For example, the inventive principle for “extraction” is described in the first specification as “extracting from an object a part or property of the product that has a negative impact”, where extraction is the functional action v, the object is the functional receptor o, and the part or property that has a negative impact is the property parameter p. This principle is expressed in the following equation [24]:

$$\mathrm{P}=[\mathrm{f}\mathrm{u}\mathrm{n}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n},\mathrm{f}\mathrm{u}\mathrm{n}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{a}\mathrm{l}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{e}\mathrm{p}\mathrm{t}\mathrm{o}\mathrm{r},\mathrm{a}\mathrm{t}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{b}\mathrm{u}\mathrm{t}\mathrm{e}\mathrm{p}\mathrm{a}\mathrm{r}\mathrm{a}\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{e}\mathrm{r}]$$

(1)

This functional semantics is applied to the categorical decomposition of inventive principles to form an interrelated knowledge space.

2.2. CRF-Based Demand Segmentation Analysis

In order to realise the calculation of the word separation analysis of the functional sessions in the FPBS process layer, and to arrive at the v + o + p normalised output results after the separation of the design requirements, this paper adopts a required word separation analysis method based on CRF (conditional random field) [25]. CRF is a Chinese word separation method based on the generation of words using word annotation. The core logic of this method is that each word used in the process of generating a word group is always in one position of the word group. There are generally four word positions in a phrase: those distributed at the beginning of the phrase (B), those in the middle of the phrase (M), those at the end of the phrase (E), and those distributed individually into words (S); then, the word separation becomes a matter of word annotation for the string sequence.

Therefore, the CRF algorithm was chosen to carry out the required split analysis calculations in this study, and the specific process is shown in Figure 2.

In order to obtain the most probable word separation solution, all the word separation solutions must be calculated; after performing the calculations, the probability of each solution is compared and the solution with the highest probability is selected as the word separation solution. In this algorithm, let X = {x₁, x₂,…, x_i} denote the observed input data sequence (input string) and Y = {y₁, y₂,…, y_i} denote the predicted state sequence (each state represents a word position token). Then, given an input string sequence, for the parameters ω = {ω₁, ω₂,…, ω_K}, the conditional probability of its output sequence of word positions is given by the following equation [25]:

$$P\left(\mathrm{y}\left|x\right.\right)=\frac{1}{Z\left(x\right)}\left({\displaystyle \sum _{i=1}^I{\displaystyle \sum _{k=1}^K{\omega }_k{f}_k\left(y_{i-1},y_i,x,i\right)}}\right)$$

(2)

where Z(x) is the normalization factor, which ensures that the sum of the conditional probabilities of all the possible word position sequences is one, and is defined as the following equation [25]:

$$Z\left(x\right)={\displaystyle \sum _y\exp }\left({\displaystyle \sum _{i=1}^I{\displaystyle \sum _{k=1}^K{\omega }_k{f}_k\left(y_{i-1},y_i,x,i\right)}}\right)$$

(3)

f_k(y_i−1,y_i,x,i) is an arbitrary feature function for expressing the possible linguistic features of a given context, and is usually a binary representation function, expressed as following equation [25]:

$$f_k\left(y_{i-1},y_i,x,i\right)=\left\{\begin{array}{c}1,condition\\ 0,others\end{array}\right.$$

(4)

The feature function of the conditional random field model can integrate all the features of the transfer feature y_i−1→y_i, and the sequence of observable words, Y, at the moment of the hidden, variable lexical position. ω_k is a parameter that is learned from the training corpus, and is the weight of the corresponding feature function f_k(y_i−1,y_i,x,i), which can have a value ranging from −∞ to +∞. Given the conditional random field model defined using Equation (2), the most probable sequence of word position tokens for any input string is given by Equation (5) [25], to find the probability maximum:

$$Y*=\arg \max P(y\left|x\right.)$$

(5)

The CRF-based word separation algorithm is used to obtain the most probable word separation solution according to Equation (5), so that the design requirements can be analysed in terms of word separation.

2.3. Word Similarity Calculation Method

Currently, one of the methods for word similarity calculation is word similarity calculation based on the use of semantic dictionaries. The dictionary used in this study is the “Extended Version of the Synonym Word Forest of the Information Retrieval Research Laboratory of Harbin Institute of Technology” published by the Harbin Institute of Technology.

The synonym word forest provides a total of five levels of coding, and the word coding table is shown in

Table 1. List of word codes.

Coding Bits	1	2	3	4	5	6	7
Examples of symbols	C	b	1	2	A	0	7
Symbolic properties	Major categories	Medium categories	Minor categories		Word group	Atomic word group
Level	Level 1	Level 2	Level 3		Level 4	Level 5

.

Tian Jule et al. [26] used the encoding and structural features of synonym word forests, combined with the similarity and relevance of words, to implement a method for calculating word similarity based on the path and depth. For two word senses, s1 and s2, the algorithm only considers the processing at the branch, and then calculates the value of similarity without considering the hierarchical numbering after stratification. Its similarity calculation, Formula (6) [25], is shown below:

$$sim\left(v_1,v\right)=init\left(v_1,v\right)\times \cos \left(n\times \frac{\pi }{180}\right)\left(\frac{n-k+1}{n}\right)$$

(6)

where init(v₁,v) is the similarity initial value function, whose independent variable is the shortest path between the sense items v₁ and v. The function takes values of 0.65, 0.8, 0.9, and 0.96 when the nearest common parent nodes of sense items v₁ and v are at levels 1, 2, 3, and 4, respectively. The expression cos(n × π/180)(n − k + 1) is the similarity initial-value adjustment parameter, n is the total number of branching-layer nodes, and k is the branch distance between the two sense terms of the nearest common parent node. In this study, v₁ is the word obtained from the analysis of the requirement syllogism, and v is the semantic syllogism representation of the 40 invention principles in the design space and the functional action words that classify the scientific effects into 36 categories. To push the invention principles is to calculate the word similarity between v₁ and all the words in v. The calculation Formula (7) [25] is shown below:

$$V*=\arg \max sim\left(v_1,v\right)$$

(7)

The similarity between the functional action and the principle of the invention is calculated, followed by a ranking of the similarity and then pushed.

2.4. Design-Knowledge Green-Property Configuration Method

A previous study [27] correlated the seven environmental efficiency elements proposed by the World Council for Sustainable Development with engineering technical parameters, and established a correlation table between the environmental efficiency elements and engineering technical parameters. On the basis of that study, Liu et al. [28] established a green design attributes and engineering technical parameters correlation table, which can easily and quickly transform green design attributes into engineering technical parameters.

The link between engineering technical parameters and the inventive principle is shown in the conflict matrix, where a pair of engineering technical parameters, “to improve” and “to deteriorate”, is usually needed to determine the recommended inventive principle. However, in many cases the designer only knows how to improve one parameter of the system, and does not know or cannot predict the corresponding contradictory parameter of the system. In those cases, the conflict matrix is not useful to help designers find the appropriate inventive principle with which to solve their design problems. Therefore, the literature [29] has found a way to recommend inventive principles corresponding to a single engineering technical parameter, by classifying the inventive principles into different classes according to the number of occurrences of each parameter in the conflict matrix, e.g., A (more than 19 occurrences), B (between 16 and 18 occurrences), C (13 to 15 occurrences), D (10 to 12 occurrences), E (7 to 9 occurrences), F (4 to 6 occurrences), and G (1 to 3 times). Those principles that occur most frequently (ranked A, B, or C in

Table 2. Table correlating green design attributes with inventive principles (partial content).

Design Principles	Design Attributes	Engineering Technology Parameter Number	A	B	C
Recycling design	Degradable/recyclable	1	35		28
		9	28, 35	13	34
		19	35, 19
	Non-toxic and non-hazardous	23	35	2	19, 7
		26	35, 3, 29	18	10
		28	32, 28, 26		3, 10
		30	22, 35, 2	1	33, 28
Disassembly design	Number of connections	1	35		28
		10	35, 10, 36	37, 18	28, 19
	Number of parts	23	10, 35, 28	18	31, 24
		26	35, 3, 29	18	10
Energy/resource efficient design	Structure	4			35
		6			18, 35
		8	35		2
		13	35	39, 2	1
		17	35, 19	2
		20
		26	35, 3, 29	18	10
	Energy/resource conversion	22	35	2	19, 7
		23	10, 35, 28	18	31, 24
Environmental emission design	Solids, liquids, and gaseous emissions	1	35		28
		23	10, 35, 28	18	31, 24
	Noise	13	35	39, 2	1
		19	35, 19

) will have a better chance of successfully solving creative design problems. Therefore, the designer can select the most frequent inventive principle based on a single engineering technology parameter and, thus, solve the design problem without the need for a contradiction analysis.

In connection with the interrelationship of environmental efficiency elements, engineering technical parameters, green design attributes, and invention principles in the above-mentioned literature, this paper establishes a correlation table for green design attributes and invention principles; taking the invention principles with grades of A, B, and C as the recommended invention principles, it completes the green attribute configuration of design knowledge (Table 2). The designer can determine the green design attributes according to the green performance requirements of the product. Afterwards, the corresponding inventive principles can be found using Table 2 and the priority of the inventive principles can be determined. These invention principles are used as green filtering results for the FPBS process design knowledge push results, so that the pushed results meet the functional requirements as well as the green performance requirements.

3. Patent Knowledge Base Based on the Principle of Invention

In summary, through the use of the green-concept-design knowledge push model for product development, designers are able to obtain the inventive principles that will meet the functional requirements as well as the green performance requirements of the product. However, as the inventive principles are derived by generalising the laws of patent problem solving, the direct pushing of the inventive principles is a relatively abstract concept for the designer, and does not satisfactorily help the designer to produce inspiring designs. Therefore, in order to complete the mapping from the principle domain to the structure domain, this paper links the inventive principles with patent cases, and builds a knowledge base of patent cases corresponding to each inventive principle. The corresponding patent cases are pushed along with the invention principles to give designers more design inspiration and, eventually, to generate conceptual design solutions.

To build a patent knowledge base using the principle of invention as the classification criterion, it is first necessary to pre-process the patent text. The key information is extracted as the basis for the classification, and the key information is analysed for its similarity with the knowledge of each principle in the inventive principles knowledge dictionary, and the inventive principle with the greatest similarity is selected as the classification label for the patent. Finally, the category, key information, and specific content of the patent are entered into the patent knowledge base, and this operation is repeated for the other patent texts to increase the data in the patent knowledge base, as shown in Figure 3.

3.1. Patent Text Pre-Processing and Classification Schemes

The abstract section of a patent is a direct distillation of the entire patent and best reflects the information about the principle of the invention and the application of the knowledge contained in the patent. In view of this, this paper adopts a pre-processing approach to extract the required information from the abstract section of the patent, i.e., the pre-processing steps include word separation, deactivation, and the extraction of key feature words from the abstract.

Among these, the feature words are the set of words that can extract a patent from a lower-level word composition and move it to a higher level with key information, according to a certain rule-based method. The feature words extracted from the patent text in this paper consist of two parts. The first part is the standard word composition of the principle knowledge space, i.e., the three feature words of functional action v + functional receptor o + attribute parameter p, which are used as the functional knowledge vector V of the patent. The second part is a feature matrix L, which is composed of several green feature vectors and is extracted as the green design attribute corresponding to the retrieved patent document.The functional knowledge vector V of a patent is represented by the following equation [24].

$$\mathrm{V}=[\mathrm{f}\mathrm{u}\mathrm{n}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n},\mathrm{f}\mathrm{u}\mathrm{n}\mathrm{c}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{a}\mathrm{l}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{e}\mathrm{p}\mathrm{t}\mathrm{o}\mathrm{r},\mathrm{a}\mathrm{t}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{b}\mathrm{u}\mathrm{t}\mathrm{e}\mathrm{p}\mathrm{a}\mathrm{r}\mathrm{a}\mathrm{m}\mathrm{e}\mathrm{t}\mathrm{e}\mathrm{r}]$$

(8)

Based on the above pre-processing process, this paper adopts the Jieba word separation algorithm to pre-process the patent abstracts, which is a word separation tool that supports the import of custom dictionaries, and can quickly and efficiently identify the words in sentences and support the extraction of keywords. It uses a combination of dynamic programming and HMM models to accurately identify the words in sentences and deal with unidentified words in the process. In summary, the invention principle knowledge dictionary, the invention principle-deactivation word dictionary, and the green design-attribute dictionary can be imported into the Jieba word separation corpus to ensure that the algorithm has a higher accuracy rate.

For the classification of the patent abstract texts, the text classification technique to be used needs to be determined. In this paper, a rule-based text classification approach is adopted. The classifier process is as follows: after pre-processing the patent text, the svop word structure similarity is calculated using the patent-classification criteria vector, V, and all the criteria constituent structures, V_i, of the principle knowledge space and, finally, the corresponding invention principle with the greatest similarity is selected as the classification label for the patent.

3.2. Patent Push Method Based on Knowledge Meta-Representation

In this paper, the ontology of each patent case is expressed by four knowledge elements, and the process for extracting and combining these four knowledge elements from a patent is shown in Figure 4. Where Cod is the coding knowledge element, Fun is the functional knowledge element, Gre is the green knowledge element, and Str is the structural knowledge element.

(1)

Coding knowledge element: This knowledge element plays a role in identifying and locating the patent. The coding rules are mainly composed of two sets of numbers, which can be regarded as a 1 × 2 vector. The first number in the vector represents the inventive principle classification of the patent to which the code corresponds, and is numbered as detailed in

Table 3. List of invention principle codes (partial content).

Number	Principle	Functional Actions	Coding
1	Splitting	Splitting	Hj30B01
2	Extraction	Extraction	Je12A12
3	Local quality	Distinguish	Dd04B01
4	Asymmetries	Add	Ih05A01
5	Portfolio	Portfolio	Ie02D01
6	Versatility	Contains	Jd06C01
7	Nested	Nested	Fa12C01
8	Weight compensation	Compensation	Ih05B01
9	Pre-action	Preset	Ka09C01
10	Precautionary measures	Preset	Ka09C01
11	Isotropic	Maintain	Jd01C01

, while the second number indicates the number of patents entered under this classification. For example, a patent coded as (1, 27) represents the 27th patent to be entered into the patent knowledge base using the “division” principle.

(2)

Functional knowledge element: The functional knowledge element represents the particular function that the patent primarily implements. Each patent in the patent knowledge base described in this paper has an inventive principle label. The process of realising a function with such an inventive principle can be summarised as the realisation process of functional action v + functional receptor o + attribute parameter p. Therefore, a functional knowledge vector, F, consisting of three feature words, functional action v + functional receptor o + attribute parameter p, can be extracted from the patent text using the Jieba syllogism as the functional knowledge element. This functional knowledge element plays the roles of accurate retrieval and pushing in the system.

(3)

Green knowledge elements: The patent knowledge push results should also meet the green needs of users. The green design keyword table is now entered into the Jieba word separation system, so that after the patent text is separated into words, a green feature word matrix can be extracted using the keywords as the green knowledge element of the patent, in order to discern whether the patent meets the green needs of the user. The green knowledge element is constructed by first constructing a green design-attribute keyword table and entering the design principle and design attribute keywords from the green design attribute table, and terms similar to them in the inventive principle knowledge dictionary, into the keyword table. At the same time, a green design-attribute keyword matrix is identified as a criterion for calculating the similarity with the green feature matrix to determine the green label of the patent; the specific content of each object in this matrix is its code in the inventor’s principles knowledge dictionary. Afterwards, the keywords are extracted using Jiaba splitting. The first extracted part is the design principle, and the keywords most similar to all the design principles are extracted as the first part of the green feature matrix; no similar words are coded as 0 in the corresponding position of the green feature matrix. The most similar keywords to all of the design attributes are extracted as the second part of the green feature matrix; no similar words are coded as 0 in the corresponding position the green feature matrix. Finally, the similarity between the standard keyword matrix and the two corresponding positions of the green feature matrix is calculated, and the one with the greatest similarity is taken as the green knowledge element of the patent instance.

(4)

Structural knowledge elements: Most of the structures in the patent cases are represented in the form of “diagram” + “interpretation”; how to extract the key diagram and its interpretation is the core problem of extracting structural knowledge elements. The analysis revealed that the interpretation of the diagram of a patent case usually explains the function and advantage of the structure. Most of the words representing functions and advantages overlap with the functional knowledge elements and green knowledge elements, as the functions implemented by the key structure are usually the core functions of the patent case. Therefore, it is possible to calculate the match between the interpretation of each schematic and the functional knowledge element, and select the structure with the highest matching level and its corresponding interpretation as the structural knowledge element of the patent case.

Using these three knowledge elements, calculated with each of the three inputs of the pushing process, allows us to conveniently realise the intelligent pushing process of patent structure knowledge. The output of the principle layer is taken as the input of the pushing process and, when calculated with the patent coding knowledge element, they mutually play the role of an initial screening of the patent library. The standard functional solution output from the functional layer and the green design-attributes output from the environment domain are treated as the second inputs of the pushing process, and are calculated along with the functional knowledge element and the green knowledge element, respectively, to ensure the accurate pushing of patent structure knowledge. After the pushing of the structural knowledge in the patent case is completed, the designer can generate design ideas for the structural improvement of the product from the invention principle and the pushed structural knowledge, and use them to complete the generation of the product concept design plan.

4. Example of a Micro-Lubrication Unit

In order to verify the validity of the above knowledge push model, this paper takes cutting processing in a particular enterprise as an example with which to carry out cutting processing using green concept design research. Cutting machining is a processing method that uses the relative motion of the cutting tool and the workpiece to remove excess metal layers from a blank. Cutting fluids play a role in cooling, lubrication, and rust prevention during this machining process, as well as playing a role in extending the service life of the tool, ensuring the machining accuracy of the product, and improving the cutting efficiency of the tool. However, the use of cutting fluids often causes serious environmental pollution, which is not conducive to the high-quality, low-consumption, high-efficiency, and green-cleanliness of the 21st century’s green manufacturing developmental direction; thus, the relevant manufacturing enterprises must develop corresponding green processing technologies. The research process entails retrieving the most matching knowledge, by mapping the design requirements onto the knowledge space and pushing out the most matching knowledge, in order to obtain a green concept design solution that meets the needs of this enterprise.

4.1. Functional Requirements Analysis

The functional requirements of the company’s cutting operations are: structural improvements to the machine tool and a reduction in cutting fluid usage. The analysis is of a functional session at the FPBS process level. The functional session takes the functional requirement as the input and the functional word separation result of the requirement as the output. The functional requirement analysis is completed using a CRF-based word-separation-analysis algorithm.

The P(y|x) values are calculated for all the cases of “reduction in cutting fluid use” as a sequence of observations, and the labelled case with the highest probability is taken as the result of the word separation. This results in a word separation of “reduction (v) + cutting fluid (o) + usage (p)”. This result is then fed into the PB principle behavioural session for word similarity matching.

4.2. Knowledge Push Results Based on Word Similarity

During the PB principle behaviour session of the FPBS process layer, a word similarity calculation is performed to match design knowledge using “reduce(V₁)” as the input. The calculation process steps are as follows:

(1)

Enter “reduce” to search in the extended glossary of synonyms and obtain the corresponding code.

(2)

Calculate the similarity of each inventive, principle functional action, V, to the input word, V₁, according to the code.

(3)

Take the corresponding word similarity as the output.

(4)

Based on $V*=\arg \max sim\left(v_1,v\right)$, the top five corresponding words with the greatest similarity are selected for pushing as the design knowledge without green filtering.

Table 4. Results of the maximum ranking of word similarity.

V	Principle of Invention	Sim
Compensation	8 Weight compensation	0.8978
Expansion and contraction	37 Thermal expansion	0.7384
Change	35 Change in physical or chemical parameters, 36 phase change	0.6769
Portfolio	5 Combinations, 40 composites	0.6769
Improve	21 Rapid action	0.6769

shows the results of the maximum ranking of word similarity.

4.3. Green Filtering of Knowledge Push Results

Then, take the user’s demand for structural improvements to the machine tool to reduce the amount of cutting fluid used. This requirement should correspond to the energy/resource saving design in the design principles in Table 3, with the design attribute structure and the engineering technical parameter 26, “amount of substance or thing”, corresponding to the inventive principle shown in

Table 5. Invention principles for green filtration.

Level	Principle of Invention
A	35 Physical or chemical parameter changes
	3 Local quality
	29 Pneumatic and hydraulic structures
B	18 Vibrations
C	10 Advance role

.

The green filtering of the inventive principles recommended in

Table 6. Information on corresponding cases in the patent knowledge base.

Number	Patents	Attachment
1	A surge filtration and adsorption-type heavy metal-contaminated soil treatment equipment [30].
2	A fully automatic pulsed micro oil and gas lubrication system [31].
3	Fluid light impurity separators [32].
4	A plate heat exchanger that changes the heat transfer efficiency by adjusting the volume [33].
5	A cyclonic atomisation nozzle for helicopter engine flushing [34].

results in a final recommendation of 35, “physical or chemical parameter changes”.

4.4. Concept Solution Generation

Next, enter the patent-case-knowledge-base system, take invention principle 35, “change in physical or chemical parameters”, as the input and check the corresponding patent cases in the patent case knowledge base; the patent information obtained is shown in Table 6.

From examining the output case information, it was found that the trace oil and gas lubrication system in patent 2, “A fully automatic pulsed trace oil and gas lubrication system” [31] and the atomising nozzle in patent 5, “A cyclonic atomising nozzle for helicopter engine flushing” [34] may be informative for reducing the amount of cutting fluid used.

Micro-lubrication is a cooling and lubrication method that uses the principle of air atomisation to change the state of the object, producing micro-sized droplets by mixing the liquid with compressed gas. Micro-lubrication devices use very small amounts of cutting fluid, which is less environmentally polluting, uses fewer resources, is efficient and energy-saving, and ensures adequate cooling and lubrication even under auxiliary conditions. In addition, this programme has the following green advantages: (1) micro-lubrication keeps the tools, workpieces, and chips outside the cutting area dry, avoiding the problem of the disposal of waste liquid; (2) micro-lubrication can be adjusted according to the working conditions regarding the optimal amount of lubricant, which can effectively eliminate the pollution caused by the suspended particles of the cutting fluid, and improve the working environment; and (3) the micro-lubrication system is simple, occupies a small amount of space, and is easily installed in a variety of machine tools.

Therefore, in response to the user’s demand for improved cutting fluid usage in cutting processes, a micro-lubrication device is used to control the use of cutting fluid as a conceptual solution. In most cases, these devices are fixed to saws, lathes, milling machines, hobbing machines, machining centres, and other machines that use externally cooled tools, as well as internally cooled machines that do not rotate at high speeds. This micro-lubrication device uses pressure atomisation. The key component is the atomising nozzle, which acts to atomise the fluid into fine particles; the basic structure of the nozzle is shown in Figure 5. In the operating mode, the liquid enters the pipeline from the inlet port and flows out from the nozzle port. Compressed air enters through the inlet and is accelerated through the narrowed orifice. In addition to the atomising nozzle, the micro-lubrication device also includes a box, an air circuit, an oil circuit, and other structures, as shown in Figure 6.

A green design was achieved while fulfilling the following user requirements: (1) provide fog cutting fluid with a certain flow rate (about 2~30 mL/h) to the cutting area, and the amount of cutting fluid can be controlled and adjusted; and (2) the spray device is connected to the air compressor, and the pressure and flow of the compressed air can be adjusted (the pressure is about 0.3~0.6 Mpa and the flow is about 1.5~3 m³/h). The overall programme form is shown in Figure 7 [35].

In addition, the use of the micro-lubrication process can extend tool life by 17~30%, which can effectively reduce the auxiliary tool change time and increase productivity by about 15%. In the target evaluation of machining costs, micro-lubrication has a clear advantage in the area of cost, due to the use of a very small amount of cutting fluid; the cost of cutting fluid is only half of what it was.

5. Conclusions

This paper is oriented towards the product concept design process and proposes a knowledge push model based on the green concept design method. Based on the SVOP functional decomposition theory of the invention principle, the invention principle knowledge space is established with the functional action as the classification standard. Based on the FPBS model, the transformation of product functional requirements using the knowledge of invention principles is achieved by adopting the split-word analysis algorithm and the word similarity algorithm. The correlation table for design attributes and invention principles is established by studying the interrelationships among the environmental efficiency elements, green design attributes, engineering technical parameters, and invention principles. Finally, the transformation from green performance requirements to invention principles is realised, so as to complete the green filtering of knowledge in the knowledge push model, and to allow the final push result to satisfy the functional and green performance requirements of the product at the same time. The knowledge push model combines traditional conceptual design methods with green design to reduce the potential negative environmental impacts of the final product. It can effectively help companies to realise the “shorter time, higher quality, greener product” approach to product development. It helps product designers to take into account both the realisation of the product’s functionality and the realisation of the product’s green attributes during the design process, ensuring that the company’s products will take a leading position in the green market in the future.

The knowledge push model still has some problems that deserve further research. The knowledge push model proposed in this paper can be further researched in terms of how to perform calculations, with the help of deep learning, with respect to demand partition analysis and word similarity calculation. In addition, because of the wide applicability of the dictionary, there are problems, such as the lack of specialised terms and the large difference between the encoding of specialised terms and their actual situation in the domain. Research on the establishment of a domain dictionary of conceptual design knowledge and for enhancing its applicability in English-speaking environments are the focus of the next steps of research.