2.1. Trends in designing and making

This section will explore trends that define the future of designing and making. Some enablers and inhibitors of these will be explored in Sections 2.3 and 2.4. Principle means of manufacturing and consumption are characterised by Global Production Networks (GPNs). Within these, the processes of design, manufacture and distribution of products span the globe in order to achieve low-cost production. These are characterised by complex supply chains and regime-driven factories, and encourage damaging environmental practices (George Reference George1986; Slade Reference Slade2006; Sumner & Mallett Reference Sumner and Mallett2013; Birtchnell & Hoyle Reference Birtchnell and Hoyle2014).

Drawbacks of GPNs came to the fore during the 2020 coronavirus pandemic, when supply chains were severed and the existing manufacturing infrastructure was unable to adapt to new demands (Gopsill et al. Reference Gopsill, Goudswaard, Snider, Johns and Hicks2021). This has added to the existing re-shoring movement that seeks to bring manufacturing ‘home’ so as not to be dependent upon global supply chains (Bailey & Propris Reference Bailey and Propris2014; Moradlou, Backhouse & Ranganathan Reference Moradlou, Backhouse and Ranganathan2017).

Distributed manufacturing is an alternative paradigm enabled by digital manufacturing technologies that can remedy issues associated with GPNs (including AM that will be explained in Section 2.3). It is characterised by a shift from large centralised manufacturing centres to diversified and distributed manufacturing resources (Wu et al. Reference Wu, Rosen, Wang and Schaefer2015) where flexibility and scalability of manufacturing units enable a move away from the centralisation tendencies of GPNs (Srai et al. Reference Srai, Kumar, Graham, Phillips, Tooze, Ford, Beecher, Raj, Gregory, Tiwari, Ravi, Neely, Shankar, Charnley and Tiwari2016).

Movements towards distributed manufacturing indicate a paradigm shift in the way things are made. The advancements in Digital Manufacturing that underpin this shift are also a pre-requisite for the paradigm of design for mass personalisation (Ozdemir, Verlinden & Cascini Reference Ozdemir, Verlinden and Cascini2022) – where products are designed to specific customer needs in a near continuous design space (Wang et al. Reference Wang, Ma, Yang and Wang2017) – in other words; designing for a market of one. In this design paradigm customers are given an active and direct role in the design process to ensure products are designed to meet their individual needs (Sikhwal & Childs Reference Sikhwal and Childs2021).

Design for mass personalisation enables leveraging the benefits of digital manufacturing technologies; however, consumers of these products remain customers of a business – that is, there is no real blurring of the consumer/producer boundary. An alternative design paradigm that leverages digital manufacturing techniques is the DoD where these barriers are fundamentally removed. This will be explored in the following section.

2.2. The DoD

Communities can be empowered to innovate for themselves due to the availability of capable and affordable manufacturing technologies such as 3D printing. If the process of designing can be made openly accessible too, this is taken a step further as communities are able to design and make products in response to local needs.

In other words, the DoD – the enabling of more non-designers to become involved in design (Fleischmann Reference Fleischmann2015) – needs to be enabled. The DoD can be broken down into three main facets: (i) increased engagement of the general public; (ii) reduction of pre-requisite technical skill and, (iii) enabling of the input of many into one goal (Goudswaard et al. Reference Goudswaard, Forbes, Kent, Snider and Hicks2019a).

In the context of this article, the first two facets are the most relevant. We can therefore define DoD as the reduction in skill level required to realise functional parts via 3D printing such that non-technical users, who typically would not be able to create parts, are able to innovate for themselves.

Parallels with the DoD can be drawn with open design (Harhoff, Henkel & Von Hippel Reference Harhoff, Henkel and Von Hippel2003; Koch & Tumer Reference Koch and Tumer2009), which has expanded from its software origins and provides a framework for sharing design information stemming from hardware as well as physical objects (Vallance, Kiani & Nayfeh Reference Vallance, Kiani and Nayfeh2001). Open design is defined as ‘the state of a design project where both the process and the sources of its output are accessible and (re)usable, by anyone and for any purpose’ (Boisseau, Omhover & Bouchard Reference Boisseau, Omhover and Bouchard2018). It is considered to be an ideal representing a direction of openness rather than something actually achievable. It allows anyone with an appropriate set of skills to innovate for themselves. It is on this point that the DoD differs as its underpinning is about reducing the requisite skills to design as a barrier to entry as opposed to in open design where a certain skillset is required in order to be able to design.

In a similar way to open design representing a direction of openness, the DoD is considered as a direction or vector towards a wider goal which may be in practice unattainable. Achieving the DoD is therefore about moving along this vector such that design practice may be more accessible to a particular audience. This shares similarities with inclusive design which seeks to reduce ‘the level of ability required to use each product, in order to improve the user experience for a broad range of customers, in a variety of situations’ (University of Cambridge Engineering Design Centre 2023).

If the aim of democratising design is to reduce the skill level required to create functional components via 3D printing, the associated difficulty in doing this is related to the manufacturing technology and available design tools for 3D printing. These will be explored in the following sections.

2.3. Material extrusion

3D printing technologies afford a wide range of advantages over traditional subtractive processes. These include permitting design freedoms not possible through other manufacturing methods (Attaran Reference Attaran2017), enabling manufacturing cost reduction (Berman Reference Berman2012) and improved sustainability outcomes (Holmström, Liotta & Chaudhuri Reference Holmström, Liotta and Chaudhuri2018). They also permit design optimisation to reduce waste, the design and manufacture of lightweight products and the manufacture of products with bespoke properties (Foresight 2015). Because of these benefits, AM technologies have been part of the home fabrication movement enabling local production of appliances, tools and replacement parts (Holzmann, Schwarz & Audretsch Reference Holzmann, Schwarz and Audretsch2018).

The most widely used 3D printing technology is MEXFootnote ¹ (Sculpteo 2021). Its applications include weather stations (Freitag Reference Freitag2015), rural farming production (Obydenkova, Anzalone & Pearce Reference Obydenkova, Anzalone and Pearce2018), disaster response (Dotz Reference Dotz2018), microscopes (Sharkey et al. Reference Sharkey, Foo, Kabla, Baumberg and Bowman2016) and PPE during the coronavirus pandemic (Rendeki et al. Reference Rendeki, Nagy, Bene, Pentek, Toth, Szanto, Told and Maroti2020). Because of the technology’s versatility, it is widely used by hobbyists and end manufacturers alike with machines available at price points from $100 right up to industrial machines for $10000 s (Pick-3D-Printer 2022).

MEX, like other AM techniques, builds parts additively layer by layer, and it is via means of a plethora of manufacturing parameters that can be controlled in this process that many of the aforementioned design freedoms and benefits are enabled. Example parameters are shown in Figure 2, including build orientation, layer height, infill percentage and number of solid shells. The design and manufacturing freedom afforded by MEX, therefore, comes at the expense of a large multi-dimensional design space.

Figure 2.

MEX manufacturing parameters.

Early applications of MEX were largely aesthetic or for prototyping, with a focus on high-quality prints to generate consistent, geometrically accurate parts with good surface finishes but with little consideration of their functional performance. As such, these parts needed to ‘look right’ rather than ‘perform right’, and because of this methods of geometric benchmarking (Rebaioli & Fassi Reference Rebaioli and Fassi2017) including an ISO standard (ISO 2019) as well as design rules (Umaras & Tsuzuki Reference Umaras and Tsuzuki2017) for MEX printing have been designated, enabling great increases in part quality with respect to geometric accuracy.

The focus is now moving towards manufacturing parts with consistent functional performance. This is due to the technology having developed further and MEX parts finding more end-use applications (Sculpteo 2021). Subsequent studies have therefore sought to characterise the mechanical performance of parts, and how this varies with respect to different manufacturing parameters. The impact that a selection of parameters has on mechanical properties is presented in Table 1. The wide-ranging impact of these parameters means that MEX and other additive manufacturing methods are heterogeneous manufacturing techniques with different printers and manufacturing parameters yielding parts with varying and anisotropic geometric and mechanical properties.

Table 1.

Impact of manufacturing parameters on mechanical properties of MEX parts

Whilst these demonstrate a range of empirical relationships, it is noted that these have taken place with a wide range of printers, polymers and conditions, meaning that generalising trends from extant data is difficult (Popescu et al. Reference Popescu, Zapciu, Amza, Baciu and Marinescu2018). In addition to these, previous work by the authors shows that the mechanical properties of parts do not scale with part size and also vary according to cross-sectional shape (Goudswaard, Hicks & Nassehi Reference Goudswaard, Hicks and Nassehi2020).

Studies into applications of these techniques in home settings have revealed that the vast majority of parts (96%) that people would wish to make with additive manufacturing technologies would be to either replicate, improve or repair existing items (Shewbridge, Hurst & Kane Reference Shewbridge, Hurst and Kane2014). In the context of Pahl and Beitz (Pahl et al. Reference Pahl, Beitz, Feldhusen and Grote2007) this corresponds to variant or adaptive design, which by their measures, combined account for 75% of design tasks. For the purpose of this article, these use cases (replicate, improve, repair, variant, and adaptive) will be collectively referred to as variant design. Variant design undertaken in the context of additive manufacturing is in itself complex as designs need to be tailored to individual printers as well as user requirements. Solution generation in these cases, therefore, requires the exploration of a multi-dimensional design space.

MEX has been shown to be a widely accessible and affordable manufacturing method that enables a wide range of design freedoms not exhibited by other manufacturing methods. This freedom is enabled by a large multi-dimensional design space underpinned by manufacturing parameters that can be individually controlled. Whilst empirical directives of these have been deduced, the impact these parameters have on mechanical performance cannot be generalised. In a general sense, it is hard to ensure the manufacture of parts with reliable mechanical performance. In the context of enabling the DoD, this is harder still as there are no design methods that can assist a non-expert designer in selecting appropriate manufacturing parameters given the functional requirements of their parts.

2.4. Design tools and techniques for additive manufacturing

This section will explore existing design tools and techniques that are currently available for additive manufacture.

The realisation of a functional part via an AM technique such as MEX commonly has five main stages:

1. 1. Identify part requirements;

2. 2. Design for AM – determine a part’s form (external geometry);

3. 3. Process planning for AM (definition of internal geometry through assignment of manufacturing parameters and tool path generation);

4. 4. Part build; and,

5. 5. Part validation.

These steps, along with their associated outputs, are shown for a traditional CAD process in Figure 3. Of these, the first two (Design for AM and Process Planning for AM) are of most interest as here design freedom can be found representing all phases of the engineering design process. Through the combination of the design of a part’s form, and Process Planning for AM, a manufactured part’s behaviour is fundamentally defined.

Figure 3.

Traditional CAD AM design process stages and respective outputs. Adapted from Qin et al. (Reference Qin, Qi, Scott and Jiang2019).

Design for AM outputs models first in a CAD tractable format which is often specific to the CAD software used. For slicing, this must be converted to a Computer-Aided Manufacturing (CAM) tractable format of which the most common is the Stereolithography (STL) file type.

An alternative to CAD is to use a design repository. By using these a user can download one of many freely available designs already in an STL format. This results in greatly reduced steps in the design process, but also limits design flexibility. These design approaches will be explored in the following sections.

2.4.4. Generative design

Generative design is about designing not only the object but a process to generate objects (Hansmeyer Reference Hansmeyer2012) or put another way, they constitute ‘design approaches that use algorithms to create designs’ (Caetano et al. Reference Caetano, Santos and Leitão2020). Via a computational exploration of design spaces, generative design techniques are able to rapidly explore a wide range of designs and have a wide range of commercial and academic applications. These existing tools can be used to fully map design spaces – requiring a user to select a candidate design, or automatically produce a candidate design based on an objective function.

Available commercial generative design packages include Fusion 360, CogniCAD, Solid Edge, Creo 7.0, MSC Apex, CATIA V6 and NX. These are capable packages for generating designs; however, they are unable to account for anisotropic material properties (as present in AM) (Junk & Burkart Reference Junk and Burkart2021). They also have a high skill barrier to entry as highly technical steps are required to, for example, define areas of material preservation, deduce load cases and highlight manufacturing constraints when setting up a generative design scenario (Buonamici et al. Reference Buonamici, Carfagni, Furferi, Volpe and Governi2020). In addition to this, accessibility of the software is reduced due to the high costs of licences – for example, Fusion 360 costs $1,200 per year plus $33 per run (Develop3D 2021).

Further applications of generative design within the context of AM can be found in the literature. These include the following:

1. 1. Methods for topological optimisation of additively manufactured parts (Silva et al. Reference Silva, Campilho, Gouveia, Pinto and Baptista2018).

2. 2. The creation of an interactive tool that generates 3D-printed legs for walking robots based on a desired motion profile (Megaro, Thomaszewski & Gross Reference Megaro, Thomaszewski and Gross2015).

3. 3. The amendment of internal and external properties of printed parts to re-distribute mass in order to allow them to balance (Prévost et al. Reference Prévost, Whiting, Lefebvre and Sorkine-Hornung2013).

4. 4. Managing manufacturing parameters to optimise the moment of inertia of a structure to increase the spinning time of a spinning top (Bächer et al. Reference Bächer, Whiting, Bickel and Sorkine-Hornung2014).

5. 5. The generation of 3D printed model joints that can provide friction during operation in order to make a functioning prototype (Calì et al. Reference Calì, Calian, Amati, Kleinberger, Steed, Kautz and Weyrich2012).

6. 6. The generation of custom 3DP infill based on required load and specific load profiles (Gopsill & Hicks Reference Gopsill and Hicks2016).

7. 7. A sketch-based method of interacting with topological optimisers for AM to simplify the interaction process (Chen et al. Reference Chen, Tao, Wang, Kang, Grossman, Coros and Hudson2018).

8. 8. A generative method based on parameterisation for the generation of variant designs in the design process (Krish Reference Krish2011).

9. 9. A bio-inspired generative design method to create bio-mimetic support structures (Zhang et al. Reference Zhang, Wang, Zhang, Gomes and Bernard2020).

10. 10. A generative approach for creating auxetic structures via MEX (Gromat et al. Reference Gromat, Gardan, Saifouni, Makke and Recho2022).

These methods are contrasted in Table 2 according to whether the methodologies presented are appropriate for non-technical users, manage internal part structures, are validated with physical parts and are applied to MEX – potential prerequisites for enabling the DoD. None of the methods covers all of these areas. Generative design approaches are therefore shown to have a wide range of applications in the context of AM and more generally for searching large design spaces but fall short of enabling the DoD for MEX. Commercial options cannot manage parts with anisotropic properties such as those manufactured via MEX. They also have high cost and skill barriers to entry much like traditional CAD packages.

Table 2.

Generative design studies in the context of AM

^a Re-casting of internal structure rather than assignment of manufacturing parameters.

Whilst they have the capability to manage large multi-dimensional design spaces, at present they have not been used in a manner that would permit a non-technical user to create functional components for MEX.

2.5. Concluding remarks and research gap

The findings of the preceding sections can be consolidated as follows. The field of Design and Manufacture is evolving towards new paradigms, moving away from the traditional approach of localised mass production to a more distributed model that focuses on creating personalised products. Whilst 3D printing technologies are widely available, design tools need to be democratised to empower people to design and manufacture themselves. In addition to this, 3D printing technologies such as MEX present challenges in terms of a large design space and mechanical properties that are hard to predict.

Traditional design tools separate the creation of form and assignment of manufacturing parameters and therefore do not explore the AM design space adequately. They are also either difficult to use (traditional CAD) or do not offer design freedoms (Design repositories). Generative design approaches are used for AM but not for MEX as they typically either do not account for the manufacturing process, do not incorporate physical testing results to validate part performance or remain difficult to use. Setting up a generative design problem from first principles is very difficult and could be considered more specialised than CAD. As a result, design repositories that host generative templates or seed designs could be a significant advantage to non-expert designers and enable the DoD.

Given the above, a research gap emerges in the application of generative design to augment the existing capabilities of design repositories. Design repositories already have a substantial user base and the incorporation of generative design could permit parts to be customised to an individual’s requirements and manufacturing resource capability whilst exploring the AM design space and leveraging the affordances of the manufacturing technology. Moreover, design repositories would also be an appropriate platform to host seed designs. These would constitute a single setup of a generative design problem that could then be run by many according to their individual requirements and constraints.

The remainder of this article will detail such a design methodology and then demonstrate and evaluate its performance.

3.1. The requirements and pillars of the DoD

The requirements of a methodology that can enable the DoD for MEX were deduced via assessment of current MEX applications, characterisations of the design for MEX process and the MEX manufacturing process itself. These are defined as follows:

1. 1. To permit a user to undertake variant design tasks: The majority of items produced via MEX are the replication, repair or improvement of existing items rather than the creation of new designs (Shewbridge et al. Reference Shewbridge, Hurst and Kane2014). If contextualised with respect to the Pahl and Beitz framework, a democratising design strategy would therefore need to cover the embodiment and detail stages (Pahl et al. Reference Pahl, Beitz, Feldhusen and Grote2007).

2. 2. To make reasoned design decisions on behalf of the user: The greatest difficulty experienced when designing for MEX is the decision-making process. This is in part due to the large and complex MEX design space (Goudswaard, Nassehi & Hicks Reference Goudswaard, Nassehi and Hicks2019b) and in-depth engineering knowledge required to create functional parts. It is these decision processes that need to be automated (Section 3.3).

3. 3. To account for variability in process and lack of process knowledge: There are significant gaps in the knowledge of the MEX manufacturing process (Huang et al. Reference Huang, Leu, Mazumder and Donmez2015). This was identified through empirical deduction of the variability of MEX components (Goudswaard et al. Reference Goudswaard, Hicks and Nassehi2020). Given this, the only means to address this is through the incorporation of feedback from physical testing in order to validate the behaviour of parts.

4. 4. To permit iterative design: Results from physical evaluation need to be incorporated and used within the design process. To allow this the design process must cater for multiple iterations.

To meet these requirements, four pillars of design democratisation are defined that can enable non-technical users to design and manufacture functional components for themselves. The pillars and how they meet the requirements above are defined in Table 3. Physical testing can be more broadly considered as functional testing of a component through its envisaged use case.

Table 3.

Pillars of design democratisation and validation

3.2. Capability profiling

Capability profiles relate to the impact that machining or manufacturing parameters have on a part’s properties. They represent the capabilities that a specific machine tool will be able to provide at a specific time on a specific product (Newman & Nassehi Reference Newman and Nassehi2009). This, combined with information about the geometry and stock material, enables a part’s characteristics to be described. This can take place at a range of levels, from the geometry of a part to chemical integration at atomic scales (Klocke, Brinksmeier & Weinert Reference Klocke, Brinksmeier and Weinert2005).

In the context of MEX, given extant progress towards geometrically accurate parts (as described in Section 2.3), a capability profile is required to relate a part’s manufacturing parameters to its mechanical properties. It is incorporated within the design methodology to enable mapping of the large solution space enabled by MEX and therefore contributes towards meeting requirement two of the DoD – making reasoned decisions on behalf of the user.

The manner in which a capability profile could be incorporated in the design for additive manufacture process is explored in depth in previous work (Goudswaard et al. Reference Goudswaard, Hicks and Nassehi2020).

The capability profile used in the design methodology incorporates five manufacturing parameters (layer height, solid shells, top/bottom layers, infill percentage and build orientation). A tensile testing program was undertaken to enable empirical deduction of the impact that each parameter had on mechanical properties. An artificial neural network was trained on the experimental data to form a capability profile able to predict a part’s ultimate tensile strength and tensile modulus based on input manufacturing parameters and geometries. The summation of design parameters multiplied by each synapse weight and hyperbolic tangent activation function enables the calculation of mechanical properties. An in-depth exploration of how this capability profile was formed and validated can be seen in Goudswaard, Hicks and Nassehi Reference Goudswaard, Hicks and Nassehi2021 along with the definition of synapse weights. Appendix A steps through how geometric and manufacturing parameters are converted into mechanical properties via the use of the capability profile.

3.3. DoD assessment methodology

The DoD seeks to increase the accessibility and involvement of non-technical users in the design process. In the context of creating functional MEX parts, this is achieved through the reduction and where possible elimination of domain-specific highly skilled decisions and operations. Given this, the assessment methodology employed is comparative, that is, characterising and contrasting the ‘as-is’ with the ‘as-democratised’ processes. Consideration of the process necessarily requires a range of design activities to be appraised including those that

1. 1. are directly CAD based;

2. 2. are related to the design cognitionFootnote ³ of the designer during the process;

3. 3. involve interaction with the artefact or its intended use environment; and

4. 4. interact with a manufacturing resource.

This section will explore existing methods of comparing design processes with respect to the above design activities before defining the approach used in this article.

3.3.2. DoD assessment process

This first stage involves logging a description of each step of the design process undertaken by a designer when generating, manufacturing and validating a component. This corresponds to element one from the literature – cognitive walk-through and step-wise design characterisation. Design steps are categorised into one of the following five bands:

1. 1. Software Interaction – for example, opening a program or exporting a file.

2. 2. Hardware Interaction – for example, operating a 3D printer.

3. 3. Decision – for example, choosing a course of action and deciding how to use the software to achieve a goal.

4. 4. Observation/Measurement – for example, evaluating a part’s performance, identifying or measuring design features, for example, the size of an interfacing component.

5. 5. Geometry alteration – generating or changing 2D or 3D geometry.

Design steps are then assigned difficulties. These are defined in Table 4 along with examples of the types of steps that could correspond to each difficulty level. Difficulty levels higher than 3 are considered as ‘specialist’, requiring knowledge acquired through, for example, an engineering degree. Values are assigned for two types of difficulty. The first regards know-how – the methods or techniques of doing something, especially something technical or practical (Collins n.d.). These relate most to step-wise design characterisation.

Table 4.

Assigned difficulties and examples

The second regards understanding – grasping ‘how a constellation of facts relevant to that subject are related to one another (causally, inferentially, explanatorily, etc.) in such a way as to be able to make new connections or draw new inferences with novel information’ (Huxster et al. Reference Huxster, Slater, Leddington, LoPiccolo, Bergman, Jones, McGlynn, Diaz, Aspinall, Bresticker and Hopkins2018). This is more related to design cognition measured via protocol analysis.

Heuristic estimation is carried out for both know-how and understanding permitting a separate comparison of each. This allows the distinction of process difficulty that is (i) directly associated with the use of a specific design tool or method (e.g. undertaking a design step) and (ii) agnostic of tools and more generally related to the design problem at hand (e.g. elucidating a course of action).

DoD assessment will be undertaken by a single designer as it is identified to be appropriate for exploratory studies in the context of protocol analysis (Hay et al. Reference Hay, Duffy, McTeague, Pidgeon, Vuletic and Grealy2017), cognitive walk-through and heuristic estimation (Nielsen Reference Nielsen1994).

Aligned with the definition of DoD presented in Section 2.2, determining whether DoD is achieved relates to the comparison of the ‘as-is’ with the ‘as-democratised’ process considering the differences between the average difficulty of process steps as well as the most difficult steps within each process. The DoD can be considered to be achieved if average and maximum step difficulties are reduced. In addition to this, this article considers a threshold difficulty of 3 as a target as this indicates a requirement of specialist knowledge to undertake the process.

The spectrum of democratisation along with a preliminary characterisation of a CAD-based design process applying the methodology outlined above is presented in Figure 4. Further details of the characterisation are reported in the literature (Goudswaard et al. Reference Goudswaard, Hicks, Gopsill and Nassehi2017) and key quantitative results for DoD comparison will be considered further in Section 5. The outcome of this previous study framed requirement 2 of DoD in Section 3.1 – to make reasoned design decisions on behalf of the user.

Figure 4.

Spectrum of democratisation for Understanding and Know-how with overall figures for CAD-based process and vector of democratisation.

4.1. Models within the GBD

Three model types are included within the GBD methodology and are as follows:

1. 1. Capability profile – to relate and convert print parameters to mechanical properties;

2. 2. Part functional model – to convert mechanical properties and geometry into part function; and,

3. 3. Structural parametric model – to convert geometric parameters into a CAM tractable STL which can then be used in slicing software.

In addition to these models, user requirements and a physical part are necessary. The process is driven by the user requirements as these are needed to direct the creation of a design instance. A physical part is necessary to validate its functional performance and direct a subsequent design iteration if it is required. Figure 5 demonstrates the models and objects within the GBD methodology and their interrelations.

Figure 5.

Models and objects in the GBD methodology, their functions and interrelations.

The capability profile (presented in Section 3.2) is specific to a manufacturing resource. The Part Functional and Structural Parametric models can be considered as seed models which can be considered as ‘the starting point for all subsequent transformations and is specifically created to allow some form of personalisation by the intended end-user/customer’ (Bingham Reference Bingham2019). In this way, the seed models represent a design space in which a design solution unique to a user’s functional requirements and manufacturing constraints can be found. The construction and navigation of this design space will be explained in the following section.

4.2. GBD methodology activities

The GBD methodology incorporates design activities from physical and digital domains and continues iteratively until a satisfactory part is manufactured. An overview of the GBD methodology is shown in Figure 6. Step 1 of the methodology involves the user inputting their functional requirements. Through simulation in the digital domain via Particle Swarm Optimisation (PSO)Footnote ⁴ and a Part Functional Model (steps 2, 3 and 4), the design of a part is predicted to meet the requirements. This part is then manufactured (step 5) and physically tested to validate its actual behaviour (step 6). This actual part behaviour is contrasted with its predicted behaviour. If the part is unsatisfactory, the physical testing results are incorporated into the next design iteration and the process is repeated until a satisfactory part is generated and validated.

Figure 6.

Overview of the generative-based design methodology (adapted from Goudswaard et al. Reference Goudswaard, Hicks and Nassehi2021b).

Figure 7 shows an IDEF0 (Ross et al. Reference Ross, Hori, Feldmann, Thornhill, Bravoco, Connor, Marca and Kornfeld1981) representation of a single iteration of the GBD methodology, demonstrating its activities and the inputs, outputs, controls and mechanisms of each.

Figure 7.

IDEF0 activity diagram for single iteration of the GBD methodology.

In the Generate instances, simulate behaviour and select instance activity (A1) User needs are translated into geometric parameters, manufacturing parameters and predicted behaviour by the mechanisms of PSO and Part function model and controls of the capability profile and part requirements. The Generate Geometric Instance (A2) activity takes geometric parameters as a control translating them to a design instance in STL format via means of the structural parametric model. In the Slice Geometry (A3) activity, the STL input is sliced in accordance with manufacturing parameters (control) by slicing software (mechanism) to generate a manufacturing instruction (output). The Manufacture Part (A4) activity part is then manufactured from raw material (input) by an MEX printer (mechanism) following a manufacturing instruction (control). In the Test Part (A5) activity, the user (mechanism) takes the manufactured part (input) and evaluates it according to testing criteria (control). The final activity, Compare part’s predicted with actual behaviour and direct search (A6), takes the actual part behaviour (input) and contrasts it with the part requirements and predicted behaviour (both controls) and generates knowledge (output) which would be used to refine subsequent design iterations.

Figure 7 shows only a single design iteration but a multi-iteration design process would take the knowledge output from A6 and use it as a control for the next iteration’s Generate Geometric Instance activity.

5.1. Defining the solution space

The solution space is defined by the seed designs. These constitute the structural parametric model, part function model and objective function. These will be defined in this section.

5.1.2. Part function model

The part function model uses mechanical properties defined by the capability profile and a static mechanics analysis of where the part is envisaged to fail.

This approach could be considered an extension to that presented by Umetani and Schmidt (Reference Umetani and Schmidt2013), which can identify weak points in designs and use this to recommend print orientations for parts. The static mechanics approach used in the following use case goes further. Through combination with a capability profile, it is able to recommend a full suite of manufacturing parameters.

The functional model, in this case, is based upon the Euler–Bernoulli beam theory as shown in equation (1). Describing the part of a curved beam in this way is valid as the radius of curvature is large compared to the cross-section.

(1) $$ \sigma =\frac{My}{I}, $$

where $ \sigma $ is equal to the UTS, $ M $ is the applied moment, $ y $ is the distance to the neutral axis and $ I $ is the second moment of area calculated about the neutral axis (as shown in equation (3)). $ \sigma $ is calculated by the artificial neural network of the capability profile as explained in Section 3.2.

Shape analysis calculates areas of both shell and infill. In this bending use case, it permits calculation of solid material at the cross-section ( $ {A}_{\mathrm{mat}} $ ) as per equation (2).

(2) $$ {A}_{\mathrm{mat}}={A}_{\mathrm{shell}}+\alpha {A}_{\mathrm{infill}}, $$

where $ {A}_x $ corresponds to shell and infill areas, respectively, and $ \alpha $ is the percentage infill.

(3) $$ {I}_{\mathrm{tot}}={I}_{\mathrm{shell}}+\alpha {I}_{\mathrm{infill}}, $$

where $ {I}_{\mathrm{tot}} $ is the second moment of area of the entire cross-section, $ {I}_{\mathrm{shell}} $ for the shell and $ {I}_{\mathrm{infill}} $ for infill.

5.1.3. Objective function

The objective function for this use case seeks to produce a part that can fulfil its required function by withstanding a given load with minimal material usage. Each design iteration also requires it to incorporate results from physical testing. The equation of the fitness function is shown in equation (4)

(4) $$ \phi =\frac{\psi L}{A_{\mathrm{mat}}}, $$

where $ \phi $ is the fitness value to be maximised, $ \psi $ is the product of penalty multipliers and $ L $ is the required load and is equal to $ \operatorname{Min}\left({F}_{\mathrm{target}},{F}_{\mathrm{required}}\right) $ . $ {F}_{\mathrm{target}} $ is the load the part needs to be able to take. $ {F}_{\mathrm{required}} $ is calculated according to equation (5)

(5) $$ {F}_{\mathrm{required}}=\tau {F}_{\mathrm{target}}, $$

where $ \tau $ is the Load Ratio and is calculated as $ \frac{F_{\mathrm{target}}}{F_{\mathrm{actual}}} $ . $ {F}_{\mathrm{actual}} $ is a result from physical testing. $ {F}_{\mathrm{target}} $ is the load that the part is required to take.

Penalty multipliers are implemented in order to do the following:

1. 1. Ensure solutions generated are within the bounds of what the capability profile can generate;

2. 2. Ensure the dimensions generated are possible (e.g. internal geometries are not larger than external geometries);

3. 3. Ensure the print is reliable; and

4. 4. Direct the algorithm more quickly to a solution.

Eight penalty multipliers were incorporated with values of either 0.1 or 0.01, these are shown in Table B1 in Appendix B.

5.2. Design outputs

Results from the iterations are shown in Table 5. The first generated component failed at a load of 130 N. This test result was fed back into the GBD to increase the target load, which then permitted the generation of a satisfactory part for the second iteration. Through the use of the methodology, a part able to meet requirements was designed in just two generations without any technical input on the part of the user. Moreover, this is achieved via rapid exploration of a seven-dimension solution space facilitating the generation of a functional part with minimal material usage.

Table 5.

Outputs from each iteration of S-hook generation

The generation of a successful component validates the functional aspect of the design methodology. As the focus of this article is on design democratisation, the next section will explore if and how design democratisation is achieved.

5.3. Evaluating the DoD

The evaluation approach defined in Section 3.3.2 was used to evaluate the difficulty of the GBD methodology contrasted with a traditional CAD-based approach.

The assigned difficulty scores and categorisation of design steps permit the appraisal of design democratisation at two levels. The first level is the overall process with attention paid to the total number of design steps, their types and the difficulty they contribute. The second level considers categorised design steps, to provide insights on where in the process difficulty has been removed. These two levels permit elucidation of the extent to which the DoD has been achieved and, if it has, the types of steps that are removed or reduced to enable this.

To facilitate the aforementioned appraisals, Table 6 presents categorised total steps for each design methodology, and average difficulty scores for both know-how (K-H) and understanding (U) in the creation of a load-bearing component. Figure 10 presents the total steps for this process at each difficulty level for both K-H and U. Figure 9 presents a breakdown of step types by category and difficulty for both K-H and U. This permits the comparison of a traditional CAD approach with the GBD methodology.

Table 6.

Overall step totals and average difficulties

Figure 9.

Generative-based democratised (GBD) methodology versus traditional CAD-based approach compared for know-how and understanding.

To provide insight into the overall process, it can be seen from Table 6 that total steps required to create a functional part are reduced from 111 to 35 with average difficulties for U and K-H reduced from 3.2 to 1.8 and 2.1 to 1.3, respectively. With respect to the reduction and/or removal of steps, geometry alteration steps are fully removed in the GBD methodology (shown in Table 6). As such, the average difficulty for geometry alteration is reduced from 3 to 0 (U) and 2.3 to 0 (K-H). Decision-making steps are reduced from 40 to 6 (shown in Table 6). Because of this, the average level of difficulty in decision-making is reduced from 3.3 to 2.2 for understanding and 3 to 0 for know-how. This demonstrates that the GBD methodology takes some design decisions on behalf of the user. Substantial reductions are also observed for other types of steps with the exception of software interaction which shows only a relatively lower decrease (30 to 24 steps).

Figure 10 shows a sum of the total steps at each difficulty level for both know-how (K-H) and understanding (U). All steps with difficulties of 4 and 5 for both K-H and U are removed for the GBD; at a difficulty level of 3, total K-H occurrences are reduced to 0, and U occurrences are reduced from 26 to 2. This can be seen by the shift down and to the left in Figure 10 when moving from trad CAD on the left to GBD on the right.

Figure 10.

Difficulty comparison for Understanding (U) and Know-how (K-H) for traditional CAD-based design versus GBD methodology.

These reductions in total design steps and average difficulties indicate that design democratisation, to some extent, has been achieved through the use of the GBD methodology. At an overall process level, this is largely due to the removal of geometry alteration steps, and a large reduction in decision-making steps. Additional mechanisms for design democratisation can be elucidated by exploring the difficulty levels of step categories in greater detail.

Comparing difficulty changes in know-how (shown in Figure 9a compared with Figure 9b) it can be seen that (i) maximum difficulty for software interaction is reduced from 3 to 2; (ii) maximum hardware interaction difficulty is reduced from 3 to 1 and (iii) maximum observation and measurement difficulty is reduced from 2 to 1.

Comparing difficulty changes in understanding (shown in Figure 9c compared with Figure 9d) it can be seen that (i) maximum decision difficulty is reduced from 5 to 3; (ii) maximum software interaction difficulty is reduced from 2 to 1.

These maximum difficulties can be considered threshold difficulty values. Their implications, along with those of the aggregate reductions in difficulty, will be considered in Section 6.

6.1. The implications of the GBD methodology on the DoD

The quantitative evaluation provided in Section 5.3 can be further explored via the difficulty levels defined in Section 3.3.2 to identify the implications of design democratisation enabled by the GBD methodology. Design steps are considered ‘specialist’ for values higher than 3 as these require knowledge that could, for example, be learned in an engineering degree. When contrasting the GBD methodology with a traditional CAD-based design approach (as shown in Section 5.3), total steps are reduced from 111 to 35 with the average difficulty of design steps shown to reduce from 3.2 to 1.8 (U) and 2.1 to 1.3 for (K-H). When using the GBD methodology, the overall process to create a functional component is shown to be shorter and less difficult than when using a traditional CAD approach. Additionally, threshold or maximum step difficulty was seen to be reduced from 5 to 3 (U) and 4 to 2 (K-H), demonstrating that all design steps in the GBD have difficulties of 3 or lower indicating that design steps considered to be specialist have been removed. Reductions in both average and threshold measures demonstrate that the DoD has been achieved. This is depicted in Figure 11 which presents the change in maximum and average difficulty values for steps in both the GBD and traditional CAD-based process. For both know-how and understanding, leftward shifts can be observed indicating that the DoD has been achieved.

Figure 11.

Difficulty comparison on the spectrum of understanding for Understanding (U) and Know-how (K-H) for traditional CAD-based design versus GBD methodology.

The DoD is about enabling non-technical users to develop parts for themselves and, in the case of the GBD methodology, a principle mechanism for enabling this is found to be the reduction of decision-making steps and removal of geometry alteration, which are undertaken on behalf of the user. However, as with any form of design democratisation, an ‘expert’ is required somewhere in the process (Goudswaard et al. Reference Goudswaard, Forbes, Kent, Snider and Hicks2019a). In the case of the GBD methodology, an expert designer is required to develop the seed models that would be hosted in the design repository. This is not indifferent to the present situation with design repositories, where users upload their designs so others may freely re-use them. In this way, the GBD methodology facilitates design with non-designers via a one-to-many relationship; an expert designer creates a seed model and (traditionally) non-designers are able to use them to create variant and bespoke parts to fulfil their needs. Steps to enable this will be considered in Section 6.3.

It is also noteworthy to consider whether cumulative (i.e. average) or threshold (i.e. maximum) difficulty is more important when enabling the DoD. The average difficulty of design steps in the process to generate a functional part could correlate to the time taken to design, where a difficulty threshold is about the requisite skill and knowledge of the designer.

6.2. Generalisability

The work undertaken in this article demonstrates the successful DoD for a load-bearing part, for which the determination of whether a part is satisfactory or not is binary and, therefore, straightforward. This represents a limitation of the illustrative application demonstrated in this article as other categories of design tasks such as fit-type problems involving interfacing components are more challenging. In these cases, it would be unclear what would need to change in order to arrive at a satisfactory design. In lieu of requiring a user to manually measure a part, automating this process would require a means of scanning the physical part to understand how it fails to meet its requirements. The GBD methodology can therefore be said to be generalisable for functional load-bearing design tasks but, for other functions such as fit, requires further work.

Related to the types of design tasks used to instantiate the design methodology is the extent to which the methodology (and implementations thereof) may reasonably be expected to capture all that is required to achieve high-performant design outputs. This can largely be considered to be dependent upon the creation of a robust framework to enable the generation of the seed models that underpin the methodology. As such, this is outside of the scope of the research presented in this article, but will be commented upon in Section 6.3.

The design process comprises a number of different areas that contribute to the difficulty. The GBD methodology enables a non-technical user, without experience in CAD, static mechanics or assignment of manufacturing parameters to generate a functional load-bearing part. In this way, it can be considered that ‘functional’ design democratisation is enabled by the GBD methodology and has been demonstrated in its instantiation by moving cognitive load from the user to the design tool. This does not, however, constitute the whole picture when it comes to the DoD. The interface which permits a user to interact with a system is also essential as even with difficulty from the aforementioned areas removed, a poorly designed interface will prevent a user from interacting with the design tool effectively. Whilst this is crucial in deploying a fully usable design tool, it was not central to the paper’s aim and would need to be developed in situ within a design repository. As such, it can be considered that the outputs and results of the article remain perfectly valid. This, along with testing with un-skilled participants, will be considered in greater detail in Section 6.3.

6.3. Further work

Difficulty reduction and subsequent design democratisation are both quantified through the undertaking of design tasks by a single user. Due to the small sample size, further user testing is necessary to confirm that the DoD is more widely achieved. As such, an avenue for further work is identified as undertaking further user testing.

In order for the method to function, expert designers are required to generate the seed designs that non-technical users require. This article has not considered the manner in which these would be generated but this is essential for enabling a practical implementation of the methodology. Further work will therefore consider the generation of a framework that expert designers would be able to follow in creating seed designs.

This article has presented a GBD methodology that could be used to augment the existing capabilities of design libraries. Following validation that the method is able to democratise design, embedding the approach into an existing design library is a logical next step. This article constitutes a proof of concept which demonstrates that the GBD methodology can achieve design democratisation. Incorporating the methodology in a design library would also enable field testing, permitting elucidation of whether the design methodology is useful and practical in its envisaged environment.