Towards the definition of assembly-oriented modular product architectures: a systematic review

During the PDP designers make decisions whose effects have a deep impact on the firm. The available techniques, such as Design for Manufacturing (DFM) and DFA (Boothroyd 1987) focus on various aspects of production. DFA is especially crucial, since assembly is cost-intensive, and is regarded as the most value-adding process (Eskilander 2001). Applying DFx (Design for X) techniques raises costs and time spent during the design phase, but such effort is more than compensated by savings in the remaining life cycle of the product, from manufacturing to supply chain, logistics, and storing (Erixon 1998; Chiu and Okudan Kremer 2011). The long-term goals of including assembly considerations in the conceptualization of design are improved product quality, increased inter-department communication and concurrent engineering mentality, and building a sustainable manufacturing system (Whitney 2004). Despite the iterative nature of PDP that allows for corrections and changes at different stages (Wynn and Eckert 2017), these goals should be targeted from the beginning (Egan 1997; Boothroyd et al. 2010).

The importance and interdisciplinarity of tools and techniques are shown in previous research (Fujita et al. 2003; Lutters et al. 2014). However, despite the benefits of DFA, they remain relatively unfamiliar in some countries, and only a small percentage of companies adopt them, even when working with concurrent design (Fujita et al. 2003). Tools used in the early design stages are less common than those used in later stages, with DFA tools for conceptual design accounting for just 6% of the total number (Chiu and Okudan Kremer 2011). Of the most commonly used DFA methodologies, including Boothroyd and Dewhurst (B&D) (Boothroyd 1987), Hitachi (Leaney and Wittenberg 1992), and Lucas (Lucas Engineering Systems Ltd 1993), only Lucas is suitable for use during the conceptual design phase (Ezpeleta Lascurain et al. 2019). A new methodology called DFA-SPDP (design for assembly-supporting product development phases), combines these tools to consider assembly throughout all stages of design, but neither this nor the Lucas methodology takes into account product modularity (Ezpeleta Lascurain et al. 2019). DFx methodologies offer several advantages for product development, among which are cost reduction, improved product quality, and faster time-to-market (Ali and Gunasekera 2023). With DFA, reducing the number of parts and simplifying the assembly brings a significant reduction in manufacturing costs, labor, and materials. Standardizing design methods aimed at ease of assembly also helps reduce defects and errors in the process, and the need for training and specialized equipment (ElMaraghy et al. 2012). Thus, the characteristics of DFA techniques suggest that their best application would be during the early stages of design.

In this context, it is relevant to introduce two fundamental concepts of assembly, namely assembly in the large and assembly in the small, which refer to different levels of product design and development. Assembly in the small refers to the technical aspects of product assembly, such as how individual parts are put together to create a functioning product. This level of assembly is primarily concerned with the physical properties and interactions of the components that make up the product. Assembly in the large, on the other hand, takes a broader perspective and considers the economic, business, and institutional issues that are involved in product design and development. In addition to the technical aspects of assembly, assembly in the large takes into account issues such as product architecture, which defines the physical relationships between the product's components and relates them to the product's functions (Whitney 2004). Suitable product architecture is essential for many important processes, from product development to the management of variety. By carefully considering the product's architecture during the design process, a company can ensure that the product is easy to assemble, cost-effective to produce, and meets the needs of its intended market. In addition, a well-designed product architecture can facilitate knowledge transfer and collaboration between different departments and stakeholders involved in the product design and development process (Ulrich et al. 2020). Integrating DFA with PAs appears as a natural step that can provide a systematic approach for evaluating the ease of assembly and disassembly of the product and deploying a flexible and correct production system from the early definition of the product's modules. This integration can lead to increased efficiency in production, reduced costs, and improved sustainability by allowing for more streamlined and flexible manufacturing processes (Bouissière et al. 2019).

A critical and often overlooked step of the PA definition process is establishing the correct decoupling point between modular product architecture—which involves breaking a product down into independent modules produced separately and combined to create a variety of products—and integral product architecture—which involves a single, unified structure and adds simplicity and reduced need for assembly, but limits design flexibility and customization (Ulrich 1995; Baldwin and Clark 2006; Lau et al. 2011). This decision impacts the entire flow of information and material, from design development to product delivery to the market (Pahl et al. 2007). Therefore, by improving the modularity of PAs it is possible to obtain many benefits, such as product differentiation (AlGeddawy and ElMaraghy 2013), flexible manufacturing, cooperation with suppliers ((Roger) Jiao et al. 2007), and creation of product families with a range of variants deriving from reusing the same components across different products (Ma and Kim 2016). These advantages lead to reduced complexity, increased product customization, and a decrease in development and production costs. An increase in product modularity has a direct, positive effect on the flow of material, lead times, work in progress, and rationalization of the factory layout (Ulrich 1995). Additionally, modularity helps the development and automation of the assembly system, as each module belongs to a separate assembly subsystem (Erixon 1998). Overall, modularity helps improve early design (Baldwin and Clark 2006), and allows greater flexibility, and cost savings in manufacturing and maintenance (Pakkanen et al. 2022), while positively impacting the assembly process efficiency (Stief et al. 2018). By following rules for ease of assembly, each module can be optimized individually, allowing designers to identify and eliminate difficulties before they become major issues (Boothroyd et al. 2010). This results in a more efficient and reliable product design, while also simplifying the assembly process and reducing the likelihood of errors (Stief et al. 2020).

The increase in product variants and customization demands has presented manufacturers with challenges in meeting customer needs while remaining competitive. Modularization has emerged as a solution to this problem, enabling manufacturers to develop a range of products or product families with increased commonality between assemblies (Baldwin and Clark 2006). It has become common practice to design a broad product range, referred to as a family, rather than focusing on a single product. To effectively design modular product families, it is crucial to consider standard interfaces and generic modules that allow for component reuse and interchangeability (Pahl et al. 2007). The design process should account for modular products from the early stages, encompassing product functionalities and interfaces between components (Allen and Carlson-Skalak 1998). Moreover, the research suggests that modularity typically arises in the early stages of product development (Fiorineschi et al. 2014). However, determining when and in which cases to leverage the benefits of modular architectures, as well as identifying the most suitable modularity types that define interfaces and interactions among modules, remains a nontrivial task (Fiorineschi and Rotini 2019). In addition to the need for standard interfaces, there are other challenges in designing successful products: e.g., balancing multiple design objectives, such as performance, cost, and user satisfaction; or incorporating feedback from stakeholders throughout the design process, which can be difficult to manage without clear communication channels and well-defined roles and responsibilities (Ulrich et al. 2020). Addressing these challenges requires careful consideration of the design methods and techniques used, as well as the organizational structure and culture surrounding the design process, thus incorporating feedback loops from users and stakeholders early and often in the design process. By doing so, manufacturers can successfully develop modular products that meet customer needs while remaining competitive in the market (Hu 2013).

One additional approach to address the defined challenges of product design is to structure a comprehensive design model that combines DFA and PA tools in the early stages of the PDP, to give designers an easy-to-follow framework for creating quality products that are easy to assemble and disassemble, highly customizable, and cost-efficient. Despite the existence of numerous techniques in the literature, the information is often limited to certain aspects of the process or spread across multiple sources, which hampers the ease of undertaking these challenges. This scattered literature also makes it tricky for readers to gain a comprehensive understanding of the topic. Hence, this paper presents a systematic literature review aimed at consolidating and presenting the information coherently. The paper provides an overview of techniques and methods that have been developed, along with their potential benefits and drawbacks. A common framework is identified to describe the structure of the models, offering designers and researchers insights into how to effectively integrate DFA and PA into their design processes. Additionally, the review sheds light on the future research direction in this field and highlights possible future areas of interest.

In a similar but not as specific nor comprehensive fashion, some recent literature reviews try to describe and frame other design methods and techniques, focusing on a variety of aspects. One focuses on the validation of the current common practice in design, arguing for the lack of a metric to measure the outcome of the design process, thus proposing strategies to objectively measure the effect of different design strategies (Eisenmann et al. 2021). A recent study acknowledged the importance of research in the early stages of production development and its impact on enabling long-term flexibility in assembly systems, given the growing request to accommodate a higher number of product variants, with shorter life cycles, and described a theoretical framework to define the phases and activities needed during the conceptual stage of development, and supported their study with multiple case studies in a leading manufacturer of heavy-duty vehicles (Svensson Harari and Fundin 2023). Another work collected all the available knowledge on design for manufacturing and assemblies (DFMA) tools and presented the associated case studies in detail (Naiju 2021), highlighting how the implementation of tools promotes collaboration and communication between stakeholders of the process. A series of papers concerned with DFMA in the building industry have been classified by past review efforts (Gao et al. 2020; Lu et al. 2021). In the context of the mechanical industry, a systematic review was conducted to investigate the enablers of transitioning manufacturing systems to the Industry 4.0 paradigm. The review aimed to describe the effects of transforming production systems into the concept of smart design engineering and its associated design process (Pereira Pessôa and Jauregui Becker 2020). Additionally, another study examined the impact of fourth industrial revolution technologies on DFMA methodologies applied to mechanical products (Formentini et al. 2022a). However, the primary objective of this review was to categorize DFMA methods based on the complexity of the analyzed products. The findings indicated that most of these methods are applied to simple products that are easy to manage and have short lead times. This highlighted the need for more comprehensive methodologies to address complexity effectively. Furthermore, the analysis of the literature revealed a decrease in research interest in this topic over the last decade. The review also emphasized the limited number of relevant DFMA cases examined during the early design phases. Wynn and Clarkson’s (2018) extended design and development process review, instead, focuses on providing an organizational framework for the literature on models of the PDP. It aims to clarify the topology of the literature, highlight different perspectives, and categorize the models based on their contexts, advantages, and limitations. The article also demonstrates how the framework integrates previous reviews and offers a new perspective on the literature. The relevance of this paper is manyfold, as it provides an overview of the different existing models and helps situate additional research on the existing academic stage. The present systematic literature review has a different scope since it aims to review and synthesize literature related to a specific part of the PDP and connecting concepts at the conceptual design phase.

To summarize, this Section provides an overview of the importance of DFA and modular PA in the product design and development process. The background described by this collection of literature highlights the key concepts and approaches that aim to achieve efficient and effective product development from the early stages. By adopting these strategies, manufacturers can improve product quality, reduce production costs, and enhance customer satisfaction. Overall, the combination of DFA and modular product architecture offers significant potential for improving the product development process.

Understanding the functional aspects of assemblies is crucial for effective modeling methodologies in the conceptual phase, making a function-oriented view of the product necessary for such frameworks. In recent years, there has been a rising number of concurrent applications of DFA methods to PA definition in the conceptual design phase (Formentini et al. 2022a). Diagrams and graphs theories are commonly used to develop models for product design and assembly processes. For example, bond − graph theory (Gui and Mäntylä 1994) is widely used in behavioral modeling, while state transition diagrams (STDs) can be combined with a library of design concepts to create a set-covering problem (Hsu et al. 1998). Another approach is to use Decision Support Problems (DSPs) based on abstracted DFA principles, which are incorporated into the conceptual stage of the development process and used to aid decision-making during the design process (Simpson et al. 1995). Configuration flow graph (CFG) and datum flow chain (DFC) graphs are also helpful in visualizing assembly processes (Kurtoglu and Campbell 2009; Demoly et al. 2011b; Stief et al. 2018, 2019). Gupta and Krishnan (1998) proposed a representation scheme for assembly sequence design based on product families, which involves grouping products with known specifications and geometries to visualize shared components. Other works on managing modularity in product families design deployed behavior-driven function − environment − structure (B-FES) modeling frameworks to graphically show the conceptual design variants (Zhang et al. 2006; Okudan Kremer et al. 2013). Overall, the use of diagrams and graph theories helps in developing models and aiding decision-making in the conceptual design and assembly process.

On the other hand, it is necessary to represent the PA, so models have been proposed, including the Product Module Reasoning System (PMRS), which mathematically calculates the feasible modules to avoid generate-and-test or heuristic search approaches. However, large combinatorial design spaces can cause issues, such as including larger spaces than just the feasible region, and not covering all life-cycle viewpoints (Rosen 1996). Matrix representation has emerged as a solution for the rationalization of modules in PAs. For example, the modularity matrix derived from the interaction and suitability matrix allows for the decomposition approach to determine modules for different products. This formal approach overcomes the issue of insufficient information, which can arise from the definition of modules when part of the needed data may not yet be specified (Huang and Kusiak 1998; Kusiak 2002). In recent years, the Design Structure Matrix (DSM) has emerged as a valuable tool for representing the interdependencies and relationships within complex systems in a concise format (Browning 2001). More recently, Blees et al. (2009), and Blees and Krause (2008), then expanded by Halfmann and Krause (2010, 2012), have a matrix representation of modularity using DSM, based on the variety of interactions in the aircraft industry. Their approach emphasizes the identification and analysis of such interactions, inherent in this domain. In addition, the requirements − system functions incidence matrix and the system functions − product components incidence matrix allow for the representation of architecture decisions. By using the mentioned matrix representations and a clustering algorithm, a new method for supporting PA design can determine modules and acquire knowledge from previously defined architectures and transfer it to other domains of interest (Pimmler and Eppinger 1994; Erixon 1998). It is noteworthy that while DSM offers several advantages, it also presents certain limitations. DSM-based modularization uses interaction patterns derived from specific contexts, which can lead to issues when applying it to different industries. Moreover, while DSM reveals the interconnections between components, it may not fully capture certain qualitative factors that influence the design of modularity. Another limitation lies in the potential complexity given by large-scale systems. This can result in excessively complex DSMs, making their creation and interpretation challenging. The effectiveness of DSM depends then on the quality of the input, which can affect the reliability of the modularized output. Despite these limitations, DSM is a valuable tool for modular design and has been pivotal in advancing modularity in various industries. Its application still needs to be approached carefully, understanding the advantages and constraints, considering the context and objectives of each project.

Other approaches to developing PAs and identifying modules have also been proposed. Stone et al. (2000b) presented an approach that starts from the product functional structure definition to analyze assemblability, creating the Energy, Material, and Signal flow model (EMS) from the Black Box model of a product’s functions and I/O flows. This approach allows for the representation of sequential function chains and the identification of modules in a design process (Stone et al. 2000a, b).

In summary, the selection of representation depends on the specific requirements and characteristics of the PA being considered. Understanding the advantages, and shortcomings of different approaches is crucial to developing effective product architectures that can be efficiently assembled.

The review starts with the development of a protocol to be followed throughout the process. Concurrently with the planning of the review process and brainstorming sessions within the research group, an initial search of the literature is performed to form the context to place this work into. Preliminary results revealed how fragmented the available references are, and that most of the studies focus on just a fraction of the topic of interest: an account of the related concepts is presented in Sect. 2 and Sect. 3. The preliminary literature search highlighted critical gaps in the DFA-PA integration field, showing the need for a deeper investigation. One of the prominent gaps is the fragmented research focus: the literature lacks a comprehensive strategy for the integration between DFA and PA, with many studies addressing only specific aspects and leaving a holistic approach to future endeavors. Additionally, there is a scarcity of cross-functional consideration across organizational domains. The collaboration and communication between design, engineering, manufacturing, and other departments are only explored at an abstract level, without actionable insights. Lastly, there is a need for clearer specification of the motivating factors behind DFA-PA, particularly understanding which stakeholders drive their integration. Building upon the preliminary analysis of the literature and the identified gaps, a set of research questions was crafted to address these gaps and develop a deeper understanding of DFA-PA integration.

To collect the body of literature to analyze, the search strategy is based on the fundamental concepts presented in Sect. 2, namely DFA and PA, and on the research questions developed in Sect. 4.1. Some disclaimers should be established: (1) the acronym “DFA” and the string “design for assembly” are synonyms for the database queries; (2) the acronym “PA” used in this manuscript does not provide any difference in the queries results if used as a synonym for “product architecture”, so it is removed; (3) the terms “product platform”, “product family”, and related strings carry a different inherent meaning to “product architecture”, so they are not included, to avoid bias in the queries results and to collect only the most relevant literature.

After trial searches to identify the best combination, the search queries are performed on the five most pertinent databases for design engineering, namely Scopus, Science Direct, Taylor & Francis, Web of Science, and Emerald Insight. The queries always follow the same structure, as in the logic equation \(\left(A\cup B\right)\cap C\), where \(A\) is the string “design for assembly”, \(B\) is the acronym DFA, and \(C\) is “product architecture”. To include all the possible relevant papers, the search is performed in all available metadata, without restricting the content to, for example, only the title and abstract. A more refined selection of papers is later completed with inclusion and exclusion criteria. The only limits applied to the search are article type (Journal or Conference), and only publication in English. No limit to the range of years is imposed. Only engineering fields are included in the query (different actions are needed for each database). The complete list of queries and database is available in Table 1.

A total of 442 articles forms the retrieved body of literature. After merging the databases, an initial step of removal of the duplicate articles reduces the number to 378. Four additional book chapters are removed (conference papers already included in the database). The general exclusion criteria specified in Table 2 are then applied, to exclude recurrent, unwanted niches of topics from the database. It is important to note that the rationale behind the exclusion of literature reviews from the database is to maintain the focus on primary studies that provide original data, analysis, and findings. Including literature reviews as primary studies could introduce duplication, and potential misinterpretation of results, as suggested in the protocol by Kitchenham (2004). Next, for the quality assessment process, the abstracts of all the remaining articles are collected and read, then a selection of only the desired articles is made, applying the specific criteria specified in Table 2. The reviewers operate reciprocally blinded, the decisions are merged, and any dispute is settled by an internal discussion. As an additional step, to enrich the temporary database, a "reference propagation" analysis is conducted. This involves assessing references both backward and forward to identify additional articles for potential inclusion. References cited by the selected articles (backward) and references that cite the selected articles (forward) are examined to ensure a thorough collection of relevant studies (only to the first level in both directions, to reduce computing effort). All the supplementary material undergoes the abstract scrutiny process and the ones that fit the inclusion criteria (or do not match the exclusion criteria) are added to the final body of literature.

At the end of the article selection, and after checking the availability of the full-text manuscripts, 90 articles are kept for analysis. A summary and recap of the study selection process is provided in Fig. 2.

The objective of this stage is to establish the data extraction form for the accurate recording of information retrieved from the collection of studies. Such information is obtained to address the previously presented research questions. In addition, the standard information from the articles is added, as presented in Table 3. The provided framework then goes through data synthesis and analysis. To perform this task and yield this review’s results, inspiration comes from relevant literature on structuring the PDP and the framework in which manufacturing companies operate. Then, a categorization is developed to classify the corpus of papers and group the results into specific clusters that give a better understanding of the aims, the goals, and the outputs of the described models. Further details on the underlying concepts of the classification protocol and the category description are presented in the following paragraphs.

Engineering design activities play a pivotal role in both technical and social fields. They draw upon technical knowledge and personal experience and creativity to affect several aspects of human life in general, and specifically, every activity within a manufacturing company. During a product life cycle, designers need to understand that the desire to make changes to a product design comes from various stakeholders, and these changes can affect either the product itself (in the way it is manufactured/assembled), an entire family of products or even, at its extreme, the manufacturers within the supply chain (Pahl et al. 2007). Hales and Gooch (2004) also highlighted that a project set in this context has multiple layers of incoming inputs and output that range from the most external, such as the environment and the market, to the entire company, to the specific activities of the manufacturing cycle. Finally, as highlighted in the comprehensive work by Wynn and Clarkson (2005), the product is the first and strongest constraint on its design process. However, other factors play an important role: for example, internal projects tend to be constrained by budget, managerial decisions, and planning. The company performing the design process typically tries to exploit the already available skills and make strategic decisions on how to develop them. Also, to use existing manufacturing resources, companies might be reluctant to commission parts that could be produced internally.

In the wake of the aforementioned considerations, a classification system categorizes the models identified in the systematic review based on the unit of analysis and context categories. This approach enables us to categorize the models based on their contribution to different areas of a company and to identify their most common objectives. The unit of analysis category indicates the model’s specific goal, which can be at the level of a product, a product family, the machines and equipment used in the production process, the production cycle itself, or the broader network of suppliers, sellers, and other activities associated with production. These five categories are named: product, family, system, cycle, and network. Also the models are classified according to their context category, which reflects the motivation behind the creation of the model, based on inputs from within and around the company. The established categories include design, manufacturing, planning, company, and market. Models falling under the design category aim to simplify the design process of the product or product family. Models under the manufacturing category are focused on refining the manufacturing process, optimizing production lines, and incorporating new manufacturing technologies. In the planning category, models strive to enhance the planning process for a product or product family by addressing issues such as capacity planning, resource allocation, or supply chain optimization. Models falling under the company category target improving the organization's functioning by standardizing processes, managing knowledge, or innovating. Finally, models under the market category are developed to address specific issues raised by clients in the market, with problems such as pricing strategies, market segmentation, or branding. The classification system which can be seen in Table 4 provides a better understanding of the models' goals and their potential applications.

Another important aspect is the of tools and methods that are used to exploit DFA concepts. As mentioned before, Chiu and Okudan Kremer (2011) classified the collected works on DfX (Design for X) tools by application provided, based on their nature, and identified five main categories:

Guidelines give directions and indications to improve the design process; checklists are to be followed to verify designs; metrics involve a combination of the two; mathematical models include equations and formulas to help with the validation; finally, methods are systematic procedures to verify the design content. In this work, the classification is expanded and revised, to cluster data from the pool of articles.

In this study, the methodology employed to investigate the integration of DFA in PA creation involves the identification of five categories of method based on the type of models described in the body of literature: guidelines, algorithms, workflows, (support) tools, and (mathematical) models. Each of these categories is further divided into specific clusters based on the final output of the model, including descriptions, matrices, graphs, alternative designs, and numeric indices. The categories appointed for this classification can be seen, along with their description, in Table 5. Guidelines refer to sets of rules or best practices that designers can follow to improve the design process. Algorithms represent a step-by-step procedure for making decisions or finding the best solution. It typically refers to a series of logical steps that are executed in a specific order to achieve a desired outcome. Algorithms are usually presented as pseudocode or actual programming code and can become quite complex. Workflows provide a visual representation of the design process, highlighting critical features or decision points. It typically includes a flowchart or diagram that shows the different stages of the process, the inputs and outputs, and the branching paths that determine the best approach. Support tools include digital or analytical aids that support designers in identifying potential DFA issues or solutions and enable them to focus on the most critical tasks. Models refer to some sort of mathematical or computational evaluation of the design process that produces a quantitative result, allowing designers to optimize the design solution. Each of these categories can in turn generate different types of output. Descriptions are qualitative evaluations of the design solution and provide directions and indications on how to improve the process. Matrices organize and display the results in an organized form, facilitating grouping and clustering. Graphs give a visual representation of the result and highlight connections between various factors. Alternative designs propose different design solutions or acceptable design solution spaces. Numeric indices provide a quantitative measure or indicator to evaluate how good a design is or to compare results.

As part of the data analysis process, to give structure to the collected selection of papers, topic modeling classification is used, a technique for representing and summarizing the contents and extracting topics from large volumes of text. Latent Dirichlet Allocation (LDA) is a generative probabilistic model for collections of discrete data (such as text corpora) (Blei et al. 2003) and is one of the most popular algorithms for topic modeling in Python.¹ So, the first step consists of collecting the abstract text data of the 90 papers and inputting them into a shareable file to upload in Python and use as a data frame. The challenge is how to extract good quality topics that are clear, segregated, and meaningful (Maier et al. 2018). This depends heavily on the quality of text preprocessing and the strategy for finding the optimal number of topics (Mimno et al. 2011). The LDA method estimates the words in the text that are likely to belong to a specific topic and categorizes them—using a predetermined number of topics (k).

To optimize the performance of the LDA algorithm, a grid search on the values of parameters alpha, beta, and k (number of topics) is performed to determine the best combination. Specifically, a series of models with varying values of alpha and beta from 0 to 1 is created and solved for k values ranging from 3 to 9 (determined as the most suitable based on some initial trials). To determine the optimal set of parameters, the coherence score for each model is the chosen deciding factor (Mimno et al. 2011). The coherence score is a measure of how semantically coherent the topics are and is used to assess the quality of each model. The combination of alpha, beta, and k that yields the highest coherence score is selected as the best set of parameters.

The best LDA model is then used to extract topics from the corpus: the resulting topics are represented as a list of words that are most relevant to each topic. Moreover, an intertopic distance map is produced to visualize the results of the modeling. The intertopic distance map shows the relationships between the topics generated by the LDA model, representing the similarity between them in a two-dimensional map. The closer two topics are to each other on the map, the more similar the corresponding topics are: it is based on a measure that considers the similarity of the keywords in each topic and their weights (Lin 1991). To further refine and make sense of these topics, topic titles (or topic IDs), descriptions, and the contribution of each topic to the overall corpus are presented, created on the basis of keywords and respective weights. The aim is to provide a complete and insightful analysis of the corpus, allowing for a deeper understanding of the topics covered in the literature.

Figure 3 shows the timeframe covered by the literature, starting in 1994 and reaching 2022 (articles retrieved in July 2022). Columns are in two shades of gray, representing the publications in indexed Journals and the articles published in Conference Proceedings. Both paper types have been considered, since the first ensures a higher quality through a tougher peer-review process than the second, whereas Conference Proceedings are important to evaluate the more recent trends in research. The two values for the corresponding year are stacked to show the total number of publications per year. During the first few years, the number of articles is not high, reaching a maximum of 4 in 2006, until 2008 when a growth in the number of publications started. The total of 8 papers published in 2020 represents the maximum. 2014 and 2015 are the only years in the last decade with just one publication each, one Journal paper (2014), and one Conference paper (2015). It must be noted that the amount for 2022 might not be the final number, given that the articles for the review were collected in July 2022. The difference between the total number of Journal papers and the total number of Conference papers is not high: the database comprises 49 journal publications and 41 conference manuscripts.

The number of publications has kept steady and low for the best part of the first decade of the new millennium. Then some interest started to grow, especially with works aiming at taking advantage of the defining heuristic methodology presented by Stone et al. (2000a). The deriving lines of research were mostly expanded from 2019 onwards when an increase of interest highlighted by a rise in the number of publications became evident. It is worth noting that most of the recently published works investigate the aerospace industry (nearly 45% of the works published from 2019), while in the previous 25 years, the interest in such complex assemblies was much lower (less than 10% of the works published before 2019). Considering the entire body of work and all the concerned industries, it can be confirmed that the addressed topic is still under investigation and novelties are still being pursued, as apparent by the number of publications in the last years, specifically in Conference Proceedings, regarded as an indication of where research is heading, where numerous articles have been issued (three, four, five publications respectively for 2019, 2020, 2021—not considering the current year—the highest numbers since 2008, notable especially considering the ongoing COVID-19 pandemic).

Figure 4 a shows a selection of the venues of publication: only the ones with the highest number of published papers from the database. Design Studies and Journal of Engineering Design are the only journals that published three or more papers, while five conferences achieved the same (over multiple years). Two of these conferences were hosted by The American Society of Mechanical Engineers, two by The Design Society, and one by CIRP Design, which had the highest number of papers published—seven in total. Figure 4b lists the most prominent authors in the field, only those who have four or more publications are shown. The low number of venues with more than two papers indicates that there is no common scientific target for this research, hindering the sharing of knowledge among research groups. However, the fact that for example Favi, Fromentini, Bouissière, and others from the same group appear multiple times in the publications suggests that the research is continuously evolving and gives internal capabilities of sharing information and knowledge. This is especially true in the aerospace industry, which has been the primary focus of these researchers in recent years.

Table 6 presents the results of the topic modeling process using LDA, which produced five main topics, identified by topic ID, description, and contribution. The five topic IDs are component representation scheme, path sampling metrics model, aircraft design analysis phase, product modular design methodology, and assembly-based product development approach. The topic descriptions provide a general overview of the research areas covered by each topic, while the topic contribution column indicates the potential impact of each topic on product design research. The latter altogether represents the general coverage of research provided by the body of literature collected here.

Adding value to the table, Fig. 5 provides a visual summary of the keywords and their weights for each topic, which aids in interpreting the results obtained from the LDA. Lastly, we present an intertopic distance map is shown in Fig. 6, generated from the LDA results, which illustrates the distribution of topics within the corpus and shows how they are related to one another based on the keywords present in each topic. In summary, these results provide an overview of the main themes and concepts found in the corpusbody of literature. It is evident from the topic modeling map that the collected works fairly represent the two main subjects of the review, DFA and PA, which we aim to address in this review. The most salient terms in the topics include words such as “design”, “product”, “assembly”, “approach”, “system”, “process”, “architecture”, and so on. Moreover, the clusters provided by the analysis show that the strongest connection is between topic 1 (assembly based product development approach) and topic 3 (product modular design methodology), which best encompass and represent the body of knowledge. They share many similar terms and are therefore considered comparable, as shown by the size of their respective bubbles and their relative position, slightly overlapping. The two topics relate to the description of product development approaches based on assembly considerations that authors have taken to improve product quality and reduce complexity. They also indicate how models try to improve the process’s efficiency by introducing modularity concepts and describing how to better design modular products. Topic 2 (aircraft design analysis) is the other main contribution within this review, as previously mentioned, due to the significant research effort posed in recent years within this industry, with many papers focusing on improving the assemblability of complex products in the design phase published from 2019 onwards. However, the topic modeling analysis also extracted two additional topics less related to the main contributions, topic 4 (component representation scheme), and topic 5 (path sampling metrics model). These topics relate to the representation of a product design, such as the visualization of connections through graphs or the organization of modules through matrices, and they describe how to evaluate the complexity of a product and the effectiveness of the design process. These are marginal topics compared to the main goals of the review but highlight the need for further work to collect and categorize research from these adjacent fields and how knowledge can be expanded in these directions.

To assess the impact of proposed models from the collected literature on various departments and aspects of manufacturing companies, the analysis of the results is based on the classification outlined in Sect. 4.4, using data from our extraction form. Figure 7 shows a bubble graph representing the distribution of papers by unit of analysis and context. The size of each bubble represents the number of papers in each combination, with larger bubbles indicating a higher number of papers. This visualization provides an overview of the distribution of research efforts across the body of literature.

Nearly half of the works focus on the product domain, which means considering and improving assembly implications for a single product, putting the emphasis on the design aspect. Product domain models appear in all five context categories: the majority (43.3%) are analyzed in the design context, followed by manufacturing (11.1%) and market (5.6%) contexts. This suggests that the design and production of products are a primary focus of analysis across various industries. Overall, more than 60% of the models proposed focus on the product domain. This is the easiest aspect to work on: an improvement in the assemblability of a single product is easy to achieve, especially by applying simple rules or a sequence of steps in a workflow. Product families are also easier to improve, but there are fewer studies focused on them. The family unit of analysis is primarily in the design (4.4%) context, with a smaller proportion in the manufacturing (2.2%) and market contexts (2.2%). Not so many studies are focused here, maybe because for academic works is easier to get a case study of a single product, even if that is part of a series of product variants, and show the effect of a specific model on one of them. On the other hand, it is more difficult to include considerations deriving from assemblability issues to the full production system layout, as per the system domain. Nevertheless, around 15% of the papers focused on improving the production system. Half of those (7.8%) are in the context of the production planning activities, the one that most closely relates to the manufacturing structure layout, followed by manufacturing (3.3%), company (2.2%), design (1.1%), and market (1.1%) contexts. The cycle unit of analysis is primarily analyzed in the company context (5.6%), followed by design (2.2%), planning (1.1%), and manufacturing (1.1%), suggesting that product life cycle analysis is primarily focused on global company processes. Finally, the network unit of analysis is primarily analyzed in the company (2.2%) context, with smaller proportions in the planning (1.1%) and market (1.1%) contexts. This shows how network analysis and the connections between companies and their markets are closely related. Overall, the analysis suggests that the design and production of individual products are the primary focus of analysis, with less emphasis on product families, systems, cycles, and networks. The context categories also vary in their focus, with planning and production processes being the primary focus of analysis.

We present in this section and Fig. 8 a second bubble chart with the horizontal x axis representing the different methods used for the models in the studies analyzed and, on the y axis, the same unit of analysis categories as before. Each bubble represents the number of papers that used a specific combination of method and unit of analysis, with the size of each bubble corresponding to the number of papers that describes that combination (bigger bubbles indicate a larger number of papers). Again, the methods considered are guidelines, workflow, tool, algorithm, and model, while the units of analysis are product, family, system, cycle, and network.

Among the methods, the workflow has the highest number of papers (32.22%), almost one-third of the papers decided to use a workflow to describe their method for product design optimization. It is most likely considered the easiest to follow, and it does not require high computational power to be executed, making it a viable process to be analogically performed by a heterogeneous design team, without processing power aid. The other methods utilized in this domain are tools (12.22%) and guidelines (6.67%), which also provide substantial assistance during the process by suggesting the correct concepts to implement and excluding undesirable solutions, but without the schematized routine of a workflow. Algorithm and mathematical model methods have a lower percentage of papers (5.56% and 4.44%, respectively), both because they are more difficult to create and formulate in the first place, and because they require more computational or knowledge effort to be implemented in real-case scenarios. In terms of units of analysis, the product category has the highest number of papers (58.89%), followed by the system (21.11%), cycle (15.56%), family (9.99%), and network (3.33%). The most studied unit of analysis is the product, with 29 papers using the workflow method, followed by 11 papers using the tool method. Overall, the results suggest that the workflow method is the most commonly suggested method in engineering design research to formulate new models, while the other categories share lower and comparable percentages of frequency.

The five categories of methods proposed before are: guidelines, workflow, support tool, algorithm, and mathematical model. Each of these models produces an output at the end of the process that can serve as suggestions for modifying the design, for giving quantitative or qualitative measures to compare alternative designs, among other examples. These outputs are classified into the categories description, matrix, graph, alternative design, and numeric index, listed in ascending order of complexity.

The workflow models are the most frequently proposed method, with 46 papers proposing this method. Numeric index outputs are the most common in this case, with 19 papers, while also 10 papers mentioned alternative design as the final output of the model. Description, graph, and matrix are also mentioned, but with lower numbers. Since workflows are easy to follow, most of them do not stop at just presenting a qualitative description or evaluation on how to improve the design and also do not stop at a matrix or graphical representation of the design, but they push forward to try and include design consideration to at least propose alternative design or the best-hypothesized solution for a specific design problem if not even specifying a calculation pattern to evaluate a quantitative measure in the form of an index that gives an exhaustive impression of the goodness of a design by quantitatively estimating it and gets also useful to compare it to possible alternative ideas. Numeric indexes from workflow methods are not only the most numerous but also are the output in percentage more frequent amongst the numeric index proposing methods (41% of workflow methods propose numeric index, while other numeric indexes are a lower percentage of their respective method categories). In the support tool category, the highest number of outputs proposed is the alternative design (35%), but it also includes several description outputs (25%). Those tools are therefore suitable for different levels of complexity of their outputs, and though they are usually based on computers, they do not often go to proposing quantitative numerical values to evaluate the design. The reason is that they want to be just support to designers, to advise and guide them without implicitly imposing improvement by calculating a supposed optimal value. Mathematical models are also developed to offer a numerical index as output, but interestingly in 60% of the cases (3 out of 5) they are used to just guide the design process. Guidelines methods usually are developed to propose alternative designs when the set of rules has been followed (60% of the outputs), while the identified algorithms have scattered outputs, covering the full range, from descriptive guidance to quantitative measures. They are usually delegated to computers to be performed, therefore the options are several (Fig. 9).

In this Section, the reader finds a summary of the entirety of the retrieved data on case studies that are extracted from the body of works. Figure 10 presents the full list. In eleven cases no example is presented. In the remaining body, the numbers show that mechanical assemblies are the most frequent case studies. Some examples include a gearbox, a desk stapler, a product family of kettles, and a centrifugal compressor with nine components. The spectrum of examined products is quite wide, though usually limited to products with a low number of parts, given the significant number of resources that a highly complex assembly demands. However, also more complex products are analyzed, as in the case of the tool-holder carousel of a CNC machine tool (Favi et al. 2018). Seventeen other instances instead involved mechanical products connected to electricity, such as an electric wok (Stone et al. 2004) and a DC vibration motor (AlGeddawy et al. 2017). Interestingly, a high percentage of case studies (15) focused on the specific high-value, high-complexity aerospace industry, in all probability sparked by industry demands for ease of assemblability and reduction of lead time. The major contributions come from the application of cDFA methodologies in this area (Bouissière et al. 2019; Favi et al. 2020; Formentini et al. 2020). Relatively small groups of examples targeted other industry fields, such as automotive (7), electrical appliances (2), military, maritime, and food (1 each).

Several case studies are developed specifically for the presented cDFA approach. Figure 11b presents a summary of the industries featured in these studies, showing prominence towards generic mechanical assemblies (a gearbox, an aesthetic cooking hood appliance, and so on). This represents the most widely used formulation of models of interest to this review. As far as the AoD approach is concerned, the majority of the case studies fall into the small mechanical assembly category as well (see Fig. 11a), though one aerospace industry example is analyzed also in this case (Demoly et al. 2011b). The single example is not representative of the interest that the aerospace industry attracts for the cDFA approach.

Nineteen out of the ninety retrieved articles presented software capabilities specifically designed to tackle some of the issues and solve mathematical problems from the described models. Gui and Mäntylä (1994) introduced the Δ software, a computing system consisting of three parts that convert functionalities into modules, create sketches, and guide designers to the best solution. The system starts with a real model including geometrical features and aims at supporting case-based reasoning for design. Another prototype system tries to prove the feasibility of the set-covering problem and to retrieve essential components of products and conform to design for assembly, by computing the DFA index and producing an automatic synthesis of design concepts through representation with the state-transition graph (Hsu et al. 1996, 1998). The Conceptual Understanding and Prototyping (CUP) by Lombeyda and Regli (1999) quickly outputs the conceptual design as a knowledge base for further refinement and builds the SBF model of the product. Zha et al. (2001b) presented a prototype assembly-oriented design expert system (AODES) that allows designers to generate, analyze and modify the product at any stage during the design process. Specifically, it enables the minimization of the number of parts and helps select the cheapest assembly technique for the product. Other product family design systems and evaluation tools were provided in the following years (Jiang and Yan 2003; Pandit and Siddique 2008). The prototype software DASER is a product assembly planning and service mode analysis tool that takes as input the PA of a product and generates the assembly and disassembly steps to estimate total assembly time. The advantage is delivering the minimum assembly and disassembly sequence for the specified parts (Yu and Li 2006). Gupta and Okudan (2008) programmed a Data Envelope Analysis (DEA) tool and developed a framework for the computerized generation of modularized conceptual designs considering DFA index calculations, starting with the functional model and the customer needs. Their work was further expanded for the simultaneous design of PA and supply chain, verifying the impact of grouping components with similar end-of-line options (Philip et al. 2012). With the same DFA index calculations and applying the interaction matrix and suitability matrix to the modularization problem, another tool helps generate potential design combinations (Chiu and Okudan 2010). Spanning many years and providing an AOD module for a PLM system (PEGASUS) and a skeleton approach, researchers provide a framework for the automatic generation of product and assembly operation structures, exportation to CAD software, and generation of assembly skeletons, to bring the benefits of concurrent engineering into product design and assembly sequence planning (ASP) stages (Demoly et al. 2009, 2011a, 2013). Other interesting approaches include a data mining tool for product clusters formation and process agglomerations in a product structure (Kretschmer et al. 2017), and an implementation through Python programming language to integrate geometry from the very beginning of the design process (GERTRUDe) (Barbedienne et al. 2019).

Figure 12 shows a summary of the programming languages, CAD programs, and other tools that are utilized to build the software presented in this Section and the works contained in this review. The reader will recognize the names of common software, programming language, and open-source relational database management system. In the other articles, information is not available, or no software capability is used.

The topic modeling analysis revealed the main topics that constitute the body of literature. These topics mostly align with what was expected, given the initial query for article retrieval, but they also shed new light on the marginal contributions that add value to this manuscript.

The topics of “Assembly-based product development approach”, and “Product modular design methodology” clearly summarize the literature, the approaches and methodologies to product development that consider assembly as a basis for designing a product. They also show that in the context of PAs, the focus is on increasing the awareness of designers about modularity and pushing towards improving the clarity of representation of modular PAs in early design stages. This can ease the definition of the modules in a product or product family, which also brings improvements in considering assemblability issues during the PDP. The option of fully integral PAs is seldom considered when the focus is on assembly and assembly system layout. In this context, modularity is deemed to be the best approach for creating PAs, as shown in several examples. One such example is the study of product families developed through platform-based design and DFMA (Emmatty and Sarmah 2012), and a second one is in the formulation of algorithms applied to groups of product variants that help designers find the optimal granularity in modularity for the underlying PA (AlGeddawy and ElMaraghy 2013) and balance assembly complexity issues (AlGeddawy et al. 2017). The first two identified topics frame the most recurring content and show the strong connection between assembly-based design (usually identified with different DFA strategies) and the modularity of PAs, which is represented with a variety of methodologies and approaches that give instruments to designers to help them achieve manufacturing objectives.

In a secondary, but still relevant, cluster, topic modeling reveals “Aircraft design analysis”, which suggests that the aerospace industry is currently the most active in applying some of the models analyzed in the preceding sections. It also highlights the opportunity to define and use such models not only for simple products with a limited number of parts but also for complex products like aircraft, which consist of an uncountable number of pieces, a plethora of assembly techniques, a multitude of joining elements of different origin, and the need for the highest possible accuracy and quality of the final result. This emphasizes the need to extend this approach to other industries and any other complex product that would benefit from the implementation of assembly considerations. Examples of the success of this approach in the aerospace industry are works on aircraft cabin integration and installation (Halfmann et al. 2010; Halfmann and Krause 2012), assessment of the manufacturability and assemblability of aircraft systems during conceptual development (Bouissière et al. 2019; Favi et al. 2020), and the formulation and assessment of design guidelines for creating PAs of aircraft products, as well as the analysis of a cDFA model applied to aerospace products (Formentini et al. 2021a, b, 2022b, c).

The remaining clusters of topics, namely “Component representation scheme” and “Path sampling metrics model”, explore other relevant issues for this research, such as representing components and modules during the conceptual phase of product design, when no geometry or tolerance has been defined. The objective is to create the primary structure of a product and display modules and component relationships that translate into interfaces in a PA definition (Stone et al. 2000b; Kusiak 2002; Yu and Li 2006). The structures developed in this topic serve as a foundation for understanding the product's structural layout and enhance knowledge transferability among stakeholders involved in the design process. Additionally, the focus is on measuring the effectiveness and demands of the design process, including the effort required by companies (financially and time-wise) and the computational power necessary for any digital support created (Morris and Steiner 2006; Oh et al. 2020; Li et al. 2022). This aids in evaluating and improving the efficiency and quality of the design process.

In the analyzed literature, other than the aerospace industry, which alone includes 15 examples of industrial case studies out of the 90 papers, most papers selected case studies based on simple mechanical products. As discussed in the first part of this Section through topic modeling analysis, the aerospace industry attracts much interest due to its unique characteristics and complexity. However, there is limited variation regarding the fields where the approaches are tested, as most studies choose simple mechanical products as examples to test their models on. Simplicity is the main connecting point as it is enough, in most cases, to demonstrate the feasibility and provide an example of how to set the model and solve the design space solution. In the future, to validate such models, applications with a higher level of complexity should be examined. The aerospace industry has shown that this is possible, with examples of highly complex products.

Another vision for the future is to expand the scope of PA and DFA models to include products that incorporate electrical components such as printed circuit boards (PCBs) and flexible circuits. The inherent complexity of these parts and their assemblability challenges make them worthy candidates for further research. When this field of research reaches sufficient maturity, the addition of electrical connectivity considerations would be a significant enhancement to existing or novel models for conceptual DFA. This is particularly relevant given the significant investments and the amount of resources spent on deploying specialized assembly systems by manufacturers of these types of products, and the high production numbers of the industry.

Software capabilities have revolutionized the engineering design world with the introduction of CAD systems and PLM management systems. It is natural to expect similar progress in the field of conceptual DFA. Recent additions to this field include the exploration of data mining techniques and the use of Python programming language. Other tools such as MySQL for databases with matrices, tables, and rules, and XML and UML for defining structured information also attract attention. While attempts have been made to transfer design concepts directly to a CAD environment for 3D modeling capabilities, such efforts have been limited to the PEGASUS AOD module for a PLM system, for example, and the more recent GERTRUDe implementation. However, research efforts dedicated to CAD software implementation have declined in recent years, with the last valuable contributions made in the early 2010s. Instead, the major interest is growing in the field of data mining, deep and reinforcement learning techniques to train and use AI models. This represents an incredibly fast-expanding field of research, though its application to conceptual DFA has been limited to a few contributions so far.

Overall, the number of models targeting DFA at the product (or product family) level is the highest because it is easier to include considerations related to assembly in the small, at the technical level, such as improving individual part design and quality, component logistics, feeding and joining, and possibly implement decisions concerning the implementation of manual or automatic assembly. A much heavier task occurs whenever attention needs to be put on the issues of assembly at the large, system or organizational level. Those include subassemblies and assembly sequence definition, involvement of personnel from different areas, line layout deployment, automation and maintenance, data and quality management, along with business-critical decisions about the model mix, market size, and production volume, outsourcing and supply chain, to name a few.

The most common situation is therefore the one where the product is the direct and unique target of design decisions and the push to have design guidelines or a workflow to ease the assembly or manufacturing process comes from designers that want support and aid to their work. This is also a starting point for making decisions about the small perspective of assembly, but this should in turn evolve into highly impactful factors at system, company, and network organizational levels. Results denote that most models reside in the area concerning the product and family units of analysis and the design and manufacturing context categories, and not so many efforts try to push the analysis forward to increase the domain of analysis. This should be the target for future implementations of already existing methodologies, or what new models developed in the future should aim at reaching. Nevertheless, even if the number of papers that concerns the categories system, cycle, and network (unit of analysis) and planning, company, and market (context) compared to the other clusters is far more limited, in these areas the most frequent combinations are system-planning and cycle-company. One such example is the work by Asadi et al. (2016) which develops the concept of similar assembly interfaces to facilitate system flexibility and reduce the inherent complexity of production planning on a mixed-product assembly line. Most often, a direct correlation occurs between the unit of analysis or domain that the model targets for implementing design decisions and the context within which the model operates, which represents where the motivation for utilizing such a model comes from. One other example comes from the study by AlGeddawy and ElMaraghy (2013), where a product family is taken as a case to further analyze the manufacturing of its product variants and define a core platform to group them according to commonalities and differentiation of parts and components.

Thus, the classification of the collected literature has exposed a gap in this niche of topics, suggesting the need for a collective effort to extend the concepts of many models presented in the review to also study the implications of re-design on the production system, product life cycle, and the network of suppliers and stakeholders surrounding manufacturing companies. Although some of the already existing DFx methodologies, such as Design for Lifecycle and design revolving around the concepts of circularity, attempt to address these issues, none are as well developed or as widely recognized as DFA, which has been evolving for at least thirty years, as confirmed by the models presented here. Therefore, the proposed additional effort should include assembly considerations to provide a comprehensive evaluation and have a deep impact on many layers of a manufacturing organization, elevating the design and development phase to greater importance and empowering it to influence decision-making from the workshop to the managerial level.

A unitary strategy that encompasses the models revolving around the topics here presented has not been established in the past. There is no collective effort to harmonize the creation of such contributions. This has been highlighted in the results of this paper, which show how scattered among the identified categories of method the results of the model are. The workflow category is the most common, but also support tools, algorithms, and guidelines received enough attention. The spread is even more evident considering the output that different models propose. There is an even distribution among the five categories, and not even amongst the several workflow contributions a best practice could be established for choosing a specific output. As mentioned, for any unit of analysis, the model that requires one to follow a workflow methodology to improve the design is the most numerous, except for the network category, which however only presents four examples. This does not imply that employing a workflow gives the most reliable or useful method. This would depend on both the stakeholders involved and the amount of effort that is required to implement it. For example, it would be a much more productive approach to create an algorithm if a computer tool is available to be utilized, as in Demoly et al. (2011c), who created a framework to define and plan assembly sequence into the preliminary design, based on a mathematical model integrating boundary conditions such as DFA rules, and the product structure definition, and then implemented it in a module for a PLM system. Therefore, the decision about how to approach the problem should be taken in most instances case by case, and it is not trivial nor straightforward to set rules or give directions for choosing it.

The same applies to the output that the model should produce. It is true that most often a numeric index, a measurable and quantitative solution gives more information and is especially useful for getting a direct correlation between output value and action to undertake, and for comparing different designs’ output to choose the best that applies. However, for some procedures, this might not be of the highest interest, and instead some form of qualitative visualization is better. For example, Stief et al. (2018) try to simplify the similarity analysis of complex products and support their clustering. To do so they propose a step-by-step procedure outputting a final matrix that represents the synthesis of two relation matrices and the result of the similarity analysis, which shows the main links between functional subassemblies. This approach serves to build product families out of similarities between products. On the other hand, additional options are the MATLAB support tool developed by Demoly et al. (2009) which produces a graph for easy subassembly choice for assembly sequence, and also the algorithm that defines a cDFA methodology to assess the efficiency of aircraft cabin concepts, and provides three different numeric indexes, three scores representing the assembly complexity of each cabin family, each cabin individual module, and the overall architecture (Formentini et al. 2022c).

This assessment highlights the lack of a standardized collective approach for constructing models within this field. Furthermore, such an approach would need to accommodate adjustments to the model's direction (context, method, output), based on a company's specific setting, environmental factors, and operational conditions. The framework presented in the next Sect. 6.4, represents an effort to establish a structure for the development of new models or the enhancement of existing ones.

The framework shown in Fig. 13 is centered around the optimization of design and assembly, which is the objective of the models. The definition of the structure is meant to offer direction for determining appropriate courses of action for engineers and designers for the optimization of design for assembly-savvy PAs. The other main assessment portion is the operational scenario, which includes the main contributions as highlighted and classified in Sect. 5.3. In summary, it consists of the identification of the correct unit of analysis to target and the context that drives the desire for an improvement of the design process. For the complete overview of the terms included in the subparts of the operational scenario framework, please refer to Sect. 4.4. Furthermore, the framework extends the definition of the operational settings by involving the occurring modeling effort, including decisions that must be made regarding the method to follow, and the desired output at the end of the process. Additional information that should be considered is the possibility of utilizing a specific software or modeling language to enhance the process, and to include examples as case studies, and proof of concepts for the proposed model. Section 5 Results and the previous part of this Section present the findings related to the elements depicted in the framework. The discussion and description of the results, as well as the answers to the research questions, are not intended to provide an exhaustive account of all the potential gaps in a broad research field. Rather, they aim to offer guidance on the future directions that need to be pursued in the development of PA formulation in the context of DFA.

Though not directly consequential to the expressed RQs, some additional remarks coming from the body of literature are listed in this subsection.

In the wake of the now famous seventeen United Nations Sustainable Development Goals (UN SDGs), and the proposed Agenda for Sustainable Development² from 2015, it is largely expected that product development also takes a part in building the future world (e.g., SDG #9: Industry, Innovation, and Infrastructure; SDG #12: Responsible Consumption and Production). However, in the full body of work of this review, only two articles briefly introduced the problem of sustainability within this context. The problem is considered in terms of carbon emission and carbon footprint deriving from design decisions, and the perspective of a greener production deriving from considerations on materials, manufacturing procedures, and assembly methods (Philip et al. 2013; Chiu et al. 2016).

Given the integrative nature of product design, another expectation is to have a parallel development both of the product itself and its manufacturing system. Specifically, assembly process planning and assembly resources are to be considered already in the early phases of design. Planning the production system for flexibility and resilience to variation in the predicted plan and product mix is pivotal to an economically viable and time-sustainable solution. Considerations are also needed for studies that focus on one singular case (which represents the majority) to extend the solution to a more general view. Nevertheless, most of the presented works focus on a specific case and do not suggest how to consider available or planned resources in the design process.

Future research directions emerge from the insights provided, indicating a need for further work to address limitations and foster continuous improvement in modular design through the integration of DFA and PA. A significant line of research involves the integration of sustainability metrics, life cycle assessment (LCA), and circular economy principles into the methodologies. This inclusion aligns with global sustainability initiatives and ensures environmental considerations during crucial design decisions. The significance and impact of such research directions emerge from the presented analysis. The incorporation of sustainability aspects resonates with global efforts toward responsible consumption, production, and innovation, as outlined by the UN SDGs. Furthermore, the inclusion of assembly process planning into early-stage design streamlines manufacturing, develops collaboration and balances the knowledge difference between production development and production (Table 7).