Design for invention: annotation of functional geometry interaction for representing novel working principles

From 2012 to 2013 worldwide patent applications grew by 9% to 2.6 million (WIPO 2014), increasing the likelihood that a designer will unwittingly use prior art. Therefore, it is understandable why at least 24% of UK companies experienced an intellectual property (IP) dispute over the past 5 years. Damages were agreed in 30% of cases and averaged £75–£115 k (Weatherall and Webster 2014), highlighting the need for greater designer awareness of potential patent infringement, which can be gained by looking at the prior art.

The basis of a UK patent includes a description of the invention plus drawings and/or CAD model images and one or more claims (UK Intellectual Property Office 2016). The patent claims are the only aspect that define the exclusive right granted to the patent holder but the other aspects support the understanding of the invention. The first claim defines the essential technical features that distinguish the invention from what is already known in the field. A patent is infringed when elements of its claims match elements of the infringing device. In assessing whether a device (or new patent) infringes with patent claims, examiners look to see if the technical features of the device match those described in the patent, primarily as set out in the claims but, importantly, supported by the patent drawings and images.

In mechanical engineering, technical drawings and computer-aided design (CAD) model images are usually pivotal in describing the technical features of a patented invention. These images can also clarify the relationships between features of the patented design, which whilst they may be covered by the claims will not be understood until the images are scrutinised. Linking the patent claims to the patent images through annotation of the images would thus improve the understanding of both and offer an improved means of searching the images.

Patent infringement, novelty and inventive step are primarily legal judgements and, therefore, use of these terms is avoided in this paper, as the method described is primarily a design tool to assist the designer to understand the working principles of prior art and thereby avoid patent infringement and/or promote invention. The conditions in which the designer will avoid conflict with patented prior art are when they are made aware of the working principles in suitably annotated patents using the method described here. The designer will then compare the working principle of their proposed design with the prior art to avoid conflict. This paper focuses on communicating the working principles of patented prior art.

This paper is not concerned with patent search and retrieval per se but rather improving awareness of the working principles of prior art through annotation of patents. However, the increasing volume of patents makes searching and analysing them not only more important but also more challenging and hence various tools have been developed. Here we are concerned with patent retrieval for the purpose of comparing patents and in the future for comparing patents with emerging designs of new devices, rather than patent analysis as used to create patent maps, networks and clusters for commercial purposes. Conventionally, in order to identify relevant prior art using a patent retrieval system the designer, or patent professional, will typically enter appropriate keywords and their semantics will have a considerable effect on the results obtained. Therefore, a single search based on the occurrence of several key words rarely captures sufficient prior art and so commercial patent retrieval systems employ text-based search methods augmented by other techniques. For example, natural language processing (NLP) with machine learning (e.g. IBM Watson SIIP platform) has been applied to patent text search, often using statistical inference to enable text search beyond keywords and attach weightings to many different possible search results. However, statistical NLP methods are semantically weak and are only able to predict with acceptable accuracy if given sufficiently large input (Cambria and White 2014). Although the NLP approach to text searching will benefit patent search, this is not the complete picture as the need for image-based approaches is becoming more important as text-based techniques are increasingly problematical (Bhatti and Hanbury 2013). However, content-based image retrieval techniques are not well-suited to patent-images, for example they exploit colour of images whereas patent images are mostly black-and-white. The requirements of a generic patent image retrieval system have been defined (Vrochidis et al. 2010), which includes a semantic-level interpretation of images not present in contemporary patent search systems (Vrochidis et al. 2012; Bhatti and Hanbury 2013). However, semantic search has been limited to the image descriptive text (Abbas et al. 2014) and other patent image retrieval research has focused on image page orientation, segmentation and low-level feature-extraction (e.g. shape) but this does not effectively capture the technical features or working principle of the design (Li et al. 2014).

Document search methods make a distinction between Navigational Search where the aim is to find a particular document, and Research Search, where the aim is to locate a number of documents relating to the search term (Guha et al. 2003). The latter is most relevant to semantic search of patents for the purpose of prior art awareness and by inference avoiding patent infringement. Semantic search is concerned not just with the occurrence of words but also with their meaning in combination with other words. Ontology, defined as specification of a conceptualisation, is frequently used to encode semantics such that they differentiate based on knowledge and relationships external to the documents being searched. External sources such as ontologies containing semantic knowledge increasingly use linked open data (LOD) where structured formal data are expressed in ontology web language. This type of semantic search is also referred to as ontology-based search and two approaches are identified (Bontcheva et al. 2014): First, human-encoded semantics, where a person encodes semantics in a machine-readable format at the time of document publishing. This typically conforms to a standard such as resource description framework (RDF) or schema that are extensions to web mark-up languages for publishing LODs. Whilst this approach is more accurate it also requires considerable human input effort. Second, Automatic Semantic Annotation, where semantics are generated automatically; the advantage being that machine-readable semantics can be generated for all documents including pre-existing but there is a loss of accuracy. Given the importance of avoiding failure to find an important relevant patent then the first approach will be taken here. More can be found on semantic search in (Bontcheva et al. 2014).

It has already been highlighted above that patent images can be searched by low-level features or by using associated text such as titles or descriptions. However, it is important to bear in mind that the visual similarity of images, as searched by patent image retrieval systems such as PATSEEK and PatMedia (Bhatti and Hanbury 2013), is not necessarily the same as similarity of working principle of the designs that the images depict. In other words, a common working principle between designs suggests to the designer that there is a clash with the prior art. This is an especially important distinction in mechanical engineering as explaining the working principle in many mechanical design patents relies heavily on illustrating how functional relationships depend on novel geometric features. Therefore, creating patent image annotations that capture working principle, based on technical features and functional interactions, will enable more accurate patent search and retrieval.

Functional representations are well-established in engineering (Rodenacker 1966; Roth et al. 1972; Koller 1973; Hubka 1982; Pahl and Beitz 1988) where design activity is viewed as the establishment of functions related to energy, materials and signals, as appropriate. In the design of complex systems, design process follows a general systematic procedure of breaking the system function down into sub-functions, known as function decomposition. A function is both the general transformative input/output relationship of a system performing a task such as heating, measuring and squeezing; and non-transformative operations such as retaining, guiding, sealing and supporting.

Functions have generally been described using uncontrolled (arbitrary) verb-noun couplets such as ‘transfer force’, ‘reduce speed’, ‘retain bearing’, ‘guide tool’, ‘seal gap’ and ‘support beam’. Form-independent methods, that typically represent the function structure only, necessitate switching between function and form-based reasoning, whereas form-dependent methods, that superimpose function structure onto physical structure, more naturally reflect the designers’ reasoning (Aurisicchio et al. 2012, 2013). A controlled vocabulary of functions called reconciled functional basis (RFB) has been broadly applied in form-independent methods (Hirtz et al. 2002). While RFB has received academic criticisms (Aurisicchio et al. 2013) it was decided to incorporate a development of it in this paper since it provides a standard format of functional representation for the purposes of our research. Form in the context of conventional form-dependent representation usually means the structure at the level of components and higher, whereas geometric features that are often the key design detail are a lower level description than the component level and comprise primitives such as edges, holes and surfaces that may combined to form the feature or a structure.

Therefore, in mechanical design the use of semantic annotation (functional representations plus text summaries) can provide insight into how a working principle is actually achieved by the interaction between geometric features.

As a form-dependent functional representation method, the function analysis diagram (FAD) uses blocks to represent device structure and arrows with labels to represent functional relations between components. However, examples of applying conventional FAD (e.g. Aurisicchio et al. 2013) are limited to product component level, whilst its capability in capturing specific novel features of geometry is unclear. FAD originated in Invention Machine Goldfire software (Devoino et al. 1997) based upon TRIZ methodology (Altshuller 1996) and was originally intended to capture the complex network of interconnected functional relationships between subsystems common in process system design, which explains why geometric detail is overlooked.

Functional geometry interaction (FGI) can be explained with reference to a simple example: the familiar gated can lid used to seal beverage cans until they are opened by the consumer. This expired patent example and others in the paper are chosen on the basis that they are familiar everyday items with working principles that will be readily understood. They also avoid any current commercial infringement controversy and allow plenty of time for any infringement cases to have appeared in the literature. A cross-section of common gated can lid extracted from a 1967 patent is illustrated in Fig. 2.

The gate is the panel of the lid (label 21) that becomes an aperture when pushed open due to fracture along a scored line in the lid material (usually aluminium); the aperture perimeter is shown in Fig. 3 (labels 20, 22, 23, 24).

Figure 4 shows the geometry of the aperture-forming gated can lid in close detail.

In the rest of the paper only the components of the FGI, namely geometric features and their functional interactions will be highlighted. A Geometric feature specified in patents will be identified by an Underline and a functional interaction between geometric features will be shown in italics. The underside protective Resin (label 33) can be ignored. The Can lid (label 12) is a single sheet of aluminium that has a Double folded edge (labels 26 and 28) defining the Aperture (label 20) and Gate panel (label 21). This means that the Gate panel (label 21) is underneath and larger than the Aperture (label 20) that will be created when the Score cut (label 29) fails, which protects the consumer from the Sharp edge. Initial fracture of the Score cut (label 29) releases the pressure within the can and after a slight pause in action the consumer will continue tearing the rest of the Gate panel (label 21) from the Can lid (label 12) along the Score cut (label 29) to produce the complete Aperture (label 20). On closer inspection of the patent it is clear how the designer achieved his design intent for the two functions of creating gate-opening and edge separation safety. The function of creating the gate-opening depends upon the functional interaction of allow separation between the Score cut (label 29) and the Gate panel (label 21), and the separate functional interaction between the Gate panel (label 21) and Spacer strip (label 27) when it is pressed by the consumer.

At the same time, edge separation safety basically depends upon a surround functional interaction between the Double-folded edge (labels 26 and 28) and both the Score cut (label 29) and Neck (label 31). The Aperture (label 20) is stiffened by the Double-folded edge (label 26 and 28), which also aids the gate-opening.

In other words, the designer has carefully made complex decisions (their design intent) about the attributes (e.g. physical effects and material characteristics) of these seemingly simple FGI in order to achieve satisfactory functions. For example, if the Neck (label 31) produced by the Score cut (label 29) is too thin then it will prematurely fail under the pressure of the beverage, and if too thick it will be too difficult for the consumer to initiate fracture in order to open the Gate panel (label 21). Similarly, the geometry of the Double-folded edge (label 26 and 28) has to be chosen by the designer to have proximity to the finger for transferring cleaving force for fracture initiation that is balanced against separation from the finger and lips so that the consumer avoids receiving cuts from the Sharp edge produced by the fracturing and tearing of the Gate panel (label 21) from the Aperture (label 20). Even for this simple example it is clear that there are several quite complex FGI that have been considered by the designer in deciding feature details of the design. These FGI determine the novel working principle, which link to the inventive step or novelty of the invention claimed by the patent.

Figure 5 shows the cross section detail of a 1952 patented can end design in which a felt Shield (label 21) protects the Raw cutting edge when the Score line (label 15) is broken, and it also serves as a Reclosure element for the can.

Figure 6 shows conventional FAD applied to the 1952 patent following the procedure outlined in (Aurisicchio et al. 2012). Features of the invention are represented in the boxes where feature names and functional interactions use the phrases stated in the patent and only useful functional interactions are shown in the figure. Important outside objects that interact with the device, e.g. the consumer, are also represented in a box. The red outline indicates an area of interest for discussion.

Figure 7 shows conventional FAD applied to the 1967 patent, described in detail in Sect. 3.2, where the red outline is an area of interest for comparison with that of Fig. 6.

When the elements contained within the red outlined areas of Figs. 6 and 7 are compared, it reveals that both designs have a functional interaction, shield and surround, respectively, between the Raw cutting edge/Score cut, respectively, and another design feature (Shield/Folded structure, respectively) that fulfils the function of edge separation safety. This comparison implies that there is potential similarity in relation to how each design provides the edge separation safety function by isolating the Sharp edge from the consumer created by breaking the Score cut. However, the fact that there is no clear conflict with prior art is indicated by direct comparison of the patent independent claims for the two cases. In Table 1 there is a comparison of the geometric features extracted from the patent independent claims (which will be found later in Figs. 9 and 11) showing that they do not conflict. Also, there is no record in the literature of any legal cases raised regarding infringement between these two cases.

Therefore, it can be seen that conventional FAD is unable to satisfactorily distinguish between the two designs (Figs. 6, 7) because it does not represent sufficient detail to avoid jumping to the wrong conclusion that the newer patent conflicts with the older patent based on the similarity of functional relationships between the key components. Consequently, FAD Plus, or FAD+, has been developed to capture working principles at a more detailed level through better representation of functional geometry interaction. It is at this level of detail that conflict of prior art can be shown to occur in mechanical engineering design, as follows.

FAD+ enhances the diagrammatic representation of mechanical inventions beyond FAD in terms of key detailed geometric features described in patent claims and images and also represents invention hierarchy. In addition, FAD+ uses functional interaction terms developed from RFB.

Information required for developing FAD+ can be gathered from words and phrases contained within patent claims that can be categorised as geometric features and functional interactions. For example, nouns describing the invention features can be classified as geometric features and verbs can be classified as functional interactions between geometric features. These terms will be expressed using RFB for the purpose of conceptualisation and standardisation. Below is demonstrated how FAD+ diagrams were produced for the two gated can patent examples. For simplicity, only the independent claim was used and the process for generating the FAD+ diagram with the designer’s input is illustrated in Fig. 8.

Considering that FAD+ might be too complex to initially present to a designer in a patent image then a text annotation, intended to be read quickly by the designer, can be used as an initial summary of the key FGI that are detailed in a hidden underlying FAD+. The patent images chosen for text annotation would most likely be those most referenced in the patent document. The text summary is generated by collecting the most referenced geometric features as presented in the FAD+ and then including their associated functional interactions and geometric features. Simple phrases are then used. As some of the patent images do not show all of the feature labels needed for the summary, additional labels are added. For example, shield opening (label 56) in Fig. 18, aperture (label 20) and gate panel (label 21) in Fig. 19.

Figure 18 shows text annotation of the 1952 gated can lid patent image highlighting the key geometric features referred to in the short summary of the working principle based on the FAD+. Figure 19 shows the annotated image of the newer 1967 gated can lid patent.

Figure 20 shows text annotation of the two newer corkscrew patents summarising their novel working principle from the underlying (hidden) FAD+.

A designer will be able to readily understand the novel working principle of each patent with the aid of these annotated patent images derived from FAD+. In practice, rather than occlude the annotated patent figure, FAD+ would initially be hidden then revealed to the designer on request similar to comments revealed by ‘hovering’ over a comment symbol in current PDF documents.

The FAD+ method for graphical representation of FGI proposed in this paper is not intended to be a patent tool that confirms inventive step or novelty, rather it aims to highlight the novel working principles and increase the designer’s awareness of relevant patent prior art and thereby avoid patent infringement for their own design.

The method described brings functional modelling into the context of comparing working principles by virtue of the graphical nodes and edges of FAD+, which enables new ways of making statistical comparisons. FAD+ also enables a rigorous transformation of unstructured natural language patent text to structured graphical representation due to the use of ontology, which will enable automated comparison in the future. Our premise is that patent novelty and inventive step as captured by patent claims, description and images relate to details of working principles that can be embodied in FGI for some mechanical engineering designs. Our main focus is on gated can lid design but we have shown relevance to other types of mechanical devices. The gated can lid designs that were compared share the same high-level working principle using FAD and the novelty of the newer design was clearly shown to be in the novel working principle of the geometric detail based on key FGI revealed by FAD+.

Representing some mechanical engineering designs in sufficient detail of geometric features and their interactions is not addressed by FAD, as its application has been limited to the component level. Developing FAD+ to capture lower-level geometric features enables it to represent the working principles of certain classes of mechanical engineering devices as indicated by the example cases. Representing the detailed feature ownership (by the use of dashed lines) is also a novel feature of FAD+ that enhances understanding of working principles. Semantic annotation of patent images summarising the working principle through combined FAD+ and text summaries offers a tangible opportunity for a new patent search and retrieval approach that could provide more accurate results. To be clear, only patents that are annotated in the way described can be searched, which will require a strategic post hoc approach for existing patents. The annotation described in this paper is focused on how to communicate existing patented solutions to the designer and thereby effectively encourage the designer not to use the working principles of those prior art solutions. We see this encouragement to think beyond prior art as the opposite to design fixation (Jansson and Smith 1991).

It has been suggested that FAD needs a better developed syntax in order to be consistent and reliable (Aurisicchio et al. 2013). Our use of domain-specific ontologies in FAD+ addresses this as discussed in the next subsection. However, IHS Goldfire software (Goldfire Technical Knowledge Discovery 2017) employs a semantic indexing technology where high-level semantic subject-action-object items are identified in a sentence for search and trend analysis purposes (Verbitsky 2004) but this is not employed at the level of FGI detail described in this paper. FAD has been shown to have value in analysing complex designs (Lee et al. 2013; Michalakoudis et al. 2014), therefore FAD+ as a simple extension of FAD, should be capable of representing complex designs. Table 8 summarises some main features of FAD and FAD+ for comparison.

We expect the level of expertise required for FAD+ to be performed is that of a mechanical engineering graduate level of design expertise with at least 2–5 years of professional experience in order to achieve proper understanding of an invention in the domains considered here. The FAD+ diagrams in the above examples took less than 15 min to identify manually from the patent independent claims and generating the text summary took less than 5 min. Adding the graphic and text annotations to the patent document using Adobe Acrobat took less than 3 min. We believe that a FAD or a FAD+ diagram will not differ significantly between professional mechanical design engineers and not differ significantly between specialists in the device domain. However, there may well be significant differences between results from general design engineers and those from design specialists within a specific field of application. We anticipate that this process would be considerably speeded up by future automation that captures basic information about existing patents from the internet, which will mean that they are then “suggested” to the designer. The designer can then be incentivised by the User Interface of a tool currently being developed (see next section) to add more information to patents of interest, such as generating a FAD+ diagram, which then belongs to his/her patent database (which could be shared). Over time the patent database grows in relation to what has been interesting. Perhaps, third parties such as consultancies and universities will populate and share similar databases. Additionally, generating a FAD+ diagram might aid the process of writing patent claims.

A design assistant tool based on FAD+ is being developed that will aim to identify potential commonality of working principle between an emerging design and existing patents and hence identify conflict with prior art that could lead to potential infringement. Therefore, design effort will be steered away from patent conflict and towards novelty by providing real-time feedback to the designer. The sets of patents to be used will be drawn from can design, and other domains are yet to be identified. The tool envisaged, identified as Design Assistant for Semantic Comparison of Intellectual Property or DASCIP (Fig. 21), will store FAD+ diagrams and text annotations of the emerging design (3D model or patent image) that express the working principle in terms of descriptions of the FGI using domain-specific knowledge base developed for this purpose.

Figure 21 illustrates an overview of DASCIP and its core components: domain-specific ontologies, patent graphical representation, text annotated patent images and a CAD system plugin.

Initially, it is envisaged that a database of patents will be accessed by conventional search. This database of patents will be used to provide information to develop patent graphical representation and domain-specific ontologies. Annotated patent images will be developed upon successful completion of these two stages and become a new annotated patent database that can be subject to new search techniques to be developed.

Patent graphical representation, domain-specific ontologies and annotated patent images form a knowledge base which contains necessary data to perform FGI interaction analysis and similar prior art identification. Details regarding development of DASCIP core components are shown in Fig. 22.

Domain-specific ontologies will be developed adopting the UPON Lite Ontology Engineering approach (De Nicola and Missikoff 2016). The database of patents will be analysed in order to collect domain-specific data. The data will be then formulated into spreadsheets, validated and improved by domain experts (Jiang et al. 2017). Then structured data will be imported into ontology software to perform contradiction analysis and iterative improvement. At the end of this stage computerised ontologies should be ready to use.

FAD+ will be used to develop a graphical patent representation. It will incorporate the capability of representing invention hierarchy, invention geometric features and functional interactions. Terms defined in domain-specific ontologies will be employed to describe the FGI of patents. FAD+ will be validated and improved iteratively through a number of case studies and then be employed to establish a database of patent functional models with FGI identified.

Patent graphical representation will be summarised by simple sentences containing key invention features and FGI in order to develop the text annotations. At the end of this stage a database of annotated patent images will be established.

Semantic annotations will be linked to an external annotated patent database using XML files (see Camba et al. 2014 for method). The patents will be searched for comparison with the emerging design based primarily on text annotations of the original patent images plus, where relevant, a graphical representation of the design depicted in the patent. Differences in the words used in annotations can be mitigated by a reference ontology mapped to appropriate terms for working principles. A statistical analysis will be performed on the degree of association identified between aspects of the emerging design and relevant patents. Effective ways of visualising the results will be explored, not limited to statistical summaries but perhaps highlighting portions of the relevant patents.

Figure 23 illustrates how DASCIP will operate within a design process to check for potential prior art conflict between the emerging design and relevant patents. It is important to note that for existing patents to be identified, they will have been annotated using the methods described in this paper.

A CAD system plugin will allow the designer to operate DASCIP seamlessly within the CAD modelling process. The prior art conflict check starts with the designer selecting the CAD design features he/she wants to check within the CAD system. Domain-Specific Ontologies will provide a list of intended FGI for the designer to choose from in order to construct a FAD+ annotation of the design features. The ontology will be used to perform conceptualisation of design terms and patent terms to enable comparison. If similarity of FGI is identified, then the potential conflicted patents will be retrieved from the database and displayed in the CAD system plugin in the form of annotated patent images. The implications for the designer are that implementing the method described will change their design practice to become more aware of relevant prior art earlier in the design process than with much of current practice. Designer understanding of relevant working principles should improve with more exposure to patented prior art.