A Rule-Based System to Promote Design for Manufacturing and Assembly in the Development of Welded Structure: Method and Tool Proposition

Welding is a consolidated technology used to manufacture/assemble large products and structures. Currently, welding design issues are tackled downstream of the 3D modeling, lacking concurrent development of design and manufacturing engineering activities. This study aims to define a method to formalize welding knowledge that can be reused as a base for the development of an engineering design platform, applying design for assembly method to assure product manufacturability and welding operations (design for welding (DFW)). The method of ontology (rule-based system) is used to translate tacit knowledge into explicit knowledge, while geometrical feature recognition with parametric modeling is adopted to couple geometrical information with the identification of welding issues. Results show how, within the design phase, manufacturing issues related to the welding operations can be identified and fixed. Two metal structures (a jack adapter of a heavy-duty prop and a lateral frame of a bracket structure) fabricated with arc welding processes were used as case studies and the following benefits were highlighted: (i) anticipation of welding issues related to the product geometry and (ii) reduction of effort and time required for the design review. In conclusion, this research moves forward toward the direction of concurrent engineering, closing the gap between design and manufacturing.


Introduction
The product development process (PDP) is a well-known engineering activity that accompanies a new service or product from its conception to the market shelf. PDP involves steps such as drafting the concept, planning the overall design, accomplishing detailed design, and prototyping [1,2]. While the first stages of the PDP consist of iterative steps enabling the designer to figure out conceptual solutions (idea generation), the last stages of the PDP are characterized by more practical activities with recursive tasks (engineering design). Computer-aided design (CAD) is considered the key tool of the embodiment design phase and, nowadays, the CAD tool combines the initial capabilities for which it was conceived (namely, to virtually create the part, display it in a 3D view environment, verify the consistency of the final assembly and quickly realize 2D engineering drawings) with the benefits deriving from the integration of the multidisciplinary design methodologies (e.g., environmental and ergonomic assessment) [3,4].
Manufacturability describes the degree to which a product can be effectively manufactured, addressing both the overall feasibility and the excess cost [5]. Sharing the same objective, assembly is a manufacturing phase in which interchangeable parts/components were linked with geometrical data retrieved from the CAD model by using object-oriented and feature-based methods, providing a specific application (i.e., automotive) [21,22]. The third method refers to the case-based approach providing models and algorithms to determine if welding is an appropriate manufacturing process for a given design or to enable the design of the weld (i.e., suitable base materials, the welding process, and filler metals process) [23][24][25].
Even though these studies look promising in the integration of DFW principles in the early phases of product development, DFW suffers from real interoperability with 3D CAD systems and DFW principles are currently being applied downstream of the 3D modeling, by following the well-known guidelines available from the literature and company's know-how (tacit internal knowledge). Thus, manufacturability issues are identified later in the design process, when many design decisions have been finalized, which forces manufacturing engineers to either request a formal design review. Iterations required by the project revision due to manufacturing and assembly issues have tremendous impacts in terms of time and rework needed, incurring greater production costs and delays in product launch, or resulting in designs that may cause significant additional costs during production. The management of information flow between members of the project team in manufacturing companies depends on the rules adopted in the companies and often varies significantly. The possibility to re-inject manufacturing knowledge within the design department of the system, especially in the case of welded structures, is a very complex task and a challenge in the modern industry. The integration of DFW with computer-aided design software can reduce redesign and control activities, which are knowledge-intensive engineering tasks, and finally, the overall project cost [26]. For design engineers with no practical welding experience, a decision support system that can provide a set of manufacturing data would be a very helpful tool, if the welding process could be contemplated from the early design stages. Thus, two main research questions arise from the analysis of the state-of-art.

1.
How welding knowledge can be translated into explicit knowledge to assist product designers during the development of 3D models of welded structures/products? 2.
What is the set of information available from the investigation of the 3D CAD model (i.e., type of feature to recognize, parameter to query) necessary to develop a CADintegrated DFW system and tool?
By acknowledging the current gap, the goal of this paper is to describe a method to formalize welding engineering knowledge (rule-based approach) that is used as a repository for the development of a CAD-integrated decision support system (software platform) for welding-compliant products. The software platform uses, as input, geometrical data retrieved by the feature analysis of the 3D CAD model (feature recognition approach) and technical aspects related to manufacturing welding operations (DFW rules) previously collected through the knowledge-based system. In this manner, it will be possible to close the gap between the design and production departments, through the creation of a knowledgebased system. The latter can translate the tacit knowledge used during the development of welded products or structures into explicit and reusable knowledge. The outcome of this system does not concern the possibility to predict the welding parameters and avoid mandatory tasks (i.e., the welding procedure qualification), but rather the possibility to deliver a consistent product design (geometrical feature) in relation to the welding process. It is worth noting that the design issues refer to the product geometrical data connected with the welding operations (i.e., the type of welding, the material to weld, the geometry and the thickness of the weld plates, the geometry of the bevel, the location of the weld beam, the dimensions for weld beam, the possibility to use intermittent versus continuous weld, and the accessibility for the weld gun and fixture equipment) and not the welding process itself (i.e., the settings of welding parameters, current and voltage, the consistency of joint design with approved standards, the levels of distortion in the base material, the selection of the wire/rod material and diameter, the number of passes, and the welding speed), which needs a qualification procedure to fulfill the code requirements. The novel contribution of this study consists in the definition of a methodological framework that can be adopted for the development of a CAD-integrated DFW tool. This paper is structured as follows. After this introduction that recalls other studies on this topic and its main limitations, Section 2 describes the overall methodology, including materials and methods used for its development. Section 3 illustrates a sample of case studies where the proposed method was applied, and Section 4 discusses results and main limitations observed by the application of this methodology. Conclusions and final remarks are reported in Section 5.

Materials and Methods
The overall system architecture (engineering software platform) of the proposed methodology is presented in Figure 1. Starting from this general picture, the main modules are described, namely, (i) the knowledge-based system and (ii) the CAD feature recognition system. According to the framework presented in Figure 1, design data (i.e., product features, feature parameters, and product manufacturing information) stored within the 3D CAD model (CAD file) is read by the CAD feature recognition system, which analyzes the model through the use of dedicated algorithms for feature extraction based on the given scheme reported in the following section (Section 2.2). Once the features and parameters are analyzed, these parameters are compared with the features and parameters stored in the DFW rules DB which are the core of the knowledge-based (KB) system. DFW rules database (DB) are the repository collecting all the welding guidelines according to the rule-based method described in the following section (Section 2.1). When a design parameter of a specific feature (i.e., the length of a triangular weld bead) is not compliant with the rule stored within the DFW rules DB (i.e., the type of the feature is not allowed for a given geometry or the value of the parameter overcomes the threshold) the feature is highlighted in the 3D CAD model and a warning (non-validated DFW rule) is provided to the designer through the DFW graphic user interface (GUI) for corrective action. The DFW GUI is a dedicated window developed to be integrated within the CAD environment (CAD plug-in), which allows displaying both results, validated and non-validated rules. The types of features to be recognized, the parameters to be checked and related thresholds for each parameter are reported in a structured repository (DfW rules DB). This database is the core element of the KB system, which can translate tacit implicit knowledge into explicit knowledge.
For clarity, the materials and methods used to develop the two systems of the engineering software platform in the following sections are presented in the KB system (Section 2.1) and the CAD feature recognition system (Section 2.2).

KB System
Welding design is considered one of the most complex engineering activities, especially when the products are subject to very challenging specifications, for instance, within the military, shipbuilding, and aircraft industries. The definition of design rules in the field of welding is a consolidated practice that originates from the results of experimental studies (academic research) and lessons learned from previous projects (industry know-how). Thus, manufacturability analysis is part of the product development process and requires the involvement of many disciplines from design to manufacturing, metallurgy, mechanics, etc. This is a challenging task due to the heterogeneous information provided by the different roles across the engineering process (designers, engineers material specialists, and technologists). In order to translate these heterogeneous rules into explicit knowledge that can be used in the development of welding-compliant products, a rule-based method was adopted. The rule-based method for developing and classifying welding design rules is based on three main pillars, namely, (i) knowledge acquisition, (ii) knowledge processing, and (iii) knowledge representation. This classification is the ontology representation of the KB system. Knowledge acquisition refers to the review of technical documents (handbooks, reports, thesis) and the investigation of the industry's best practices for the collection of heterogeneous DFW design rules. This phase consists of two main tasks-(i) the collection of unstructured design rules for several manufacturing and assembly technologies (string data) and (ii) the identification of geometric entities (CAD features) and numerical parameters and threshold values involved in the design rules (numerical data). Knowledge processing refers to the link between the DFW rules previously collected during the knowledge acquisition phase and the geometric features of a virtual 3D model (CAD file). This phase is an essential task for transforming a tacit knowledge (DFW rules list) into a systematic design review of the product (explicit knowledge). Knowledge representation refers to the definition of a structured repository for the collection and the formalization of welding knowledge. This phase encompasses the logical definition of DFW rules and guidelines (syntax) and related information (e.g., suggestions about design changes to guarantee product weldability). It is important to underline that each time new knowledge (i.e., a design rule) is available, this rule can be recorded to be subsequently added to the repository. This is an iterative process that allows the collection of manufacturing knowledge, each time when new knowledge becomes available.

Knowledge Acquisition Phase
The knowledge acquisition phase begins by analyzing the available documents (e.g., book, research papers, technical reports, master/PhD thesis) related to the DFW topic in which tacit and unstructured knowledge is stored. In particular, in this research, the following handbooks were investigated and are reported in the references section: (i) Orlov (1976;1977) [27]; (ii) Bralla (1999) [28]; (iii) Caimbrone (2007) [29]; (iv) Poli (2001) [30]; (v) Molloy et al. (1998) [31]; (vi) El-Wakil (2019) [32]; and (vii) Boothroyd et al. (2010) [33]. It is worth noting that, in some handbooks, DFW rules are already available as a list of actions describing what to do and what to avoid during the design phase of a mechanical component to be produced using a specific manufacturing technology (i.e., the works of Orlov (1976;1977) [27]). On the other hand, according to other authors, the DFW rules are not explicitly stated, and a more in-depth analysis is necessary to extract applicable design rules. With the same aim and following the same approach, technical reports from manufacturing industries and thesis were analyzed to retrieve DFW rules. Another essential source for the acquisition of welding rules is the available documentation of commercial tools developed for (Design for Manufacturing/Design for Assembly) DFM/DFA analysis (i.e., DFMA ® tool from BOOTHROYD DEWHURST, Inc. and DFMPro ® from HCL Technologies Ltd., Mumbai, India). In addition to this, various authors organized meetings in design departments of manufacturing companies to gather the best practices and rules dedicated to given manufacturing technologies.

Knowledge Processing Phase
The knowledge processing phase begins with the definition and classification of DFW rules associated with given manufacturing technology. This phase aims to create an ontology (i.e., the structuring and formalization of data into hierarchies and classes to establish the relations among the data required for efficient machine processing that is a comprehensive description of the domain of interest (welding design rules) concerning the users' needs. Table 1 presents the overall structure of the repository used for collecting and storing the rule-related information. The structure of the repository is the semantic (logic) used to switch from tacit knowledge (unstructured) to explicit knowledge (structured). The repository stores rules based on the rule number, which is a positive, progressive number. It contains three clusters of information, namely, (i) manufacturing technology, recalling the technological aspects (i.e., manufacturing technology class and manufacturing process type) related to a given rule; (ii) material, providing material information (i.e., material class and material type) of a given rule; and (iii) CAD feature recognition, identifying geometric parameters and thresholds associated to a given rule. In addition to these sections, information about rule type is stored (i.e., info, warning, critical). An example for the W025 DFW rule is reported to clarify some aspects of the information to insert in the given structure.
Concerning the first section of the knowledge processing phase, classification of manufacturing technologies requires the definition of different clusters: (i) welding technology class (i.e., welding technology-fusion welding, solid-state welding and brazing, and soldering); (ii) manufacturing technology type-level I (e.g., arc welding, oxyfuel gas welding, resistance welding, and newer welding for the fusion welding technology class); and (iii) manufacturing technology type-level II (e.g., metal inert gas, tungsten inert gas, fluxed core, submerged arc, etc. for the arc welding manufacturing technology type-level I). The overall classification of welding technologies is proposed in Figure 2. It is worth noting that in the current research, only the fusion welding class was considered with the possibility to further extend this study to the other classes (i.e., brazing and soldering, and solid-state welding).  The adoption of these clusters is necessary to classify DFW rules that are generic for a technology class (e.g., fusion welding) or specific for a manufacturing operation of the defined technology class (e.g., metal inert gas). Indeed, a DFW rule may be valid for the generic manufacturing technology class (e.g., fusion welding) regardless of the specific process. In this case, the two levels or only the second level related to the welding technology type are not specified (N.A.; not applicable). Conversely, a welding rule may be valid only for a specific process (e.g., metal inert gas) and cannot be generalized for the welding technology class that contains this process (e.g., fusion welding).
Regarding the second section of the knowledge processing phase, classification of the material requires the definition of two clusters according to Ashby [34] classification, which is (i) material class (e.g., carbon steel, stainless steel, aluminum alloy) and (ii) material type (e.g., AISI 304, 34NiCrMo16). These two groups allow for the allocation of a given DFW rule to a generic class (e.g., stainless steel) or a specific type (e.g., AISI 304) of materials. The identification of these two clusters allows for the classification of DFW rules that are valid for any material (N.A.; not applicable), for a given material class (e.g., stainless steel), or for a given material type (e.g., AISI 304). It is worth noting that the material in the case of welding processes refers to the material of the base plates (that can be two or more different materials based on the type and the number of the base plates) and the material of the filler material (if applicable).
Regarding the third and last section of the knowledge processing phase, the classification of geometrical parameters and thresholds deals with 3D CAD features to recognize in relation to a given DFW rule. Authors defined four clusters, namely, (i) 3D CAD features (e.g., plates orientation); (ii) product manufacturing information (PMI) (e.g., roughness, tolerances); (iii) dimension/geometry (e.g., plate thickness, plate length, plate width); and (iv) rule/s to verify (e.g., thickness variation less than 1.5 times). Feature recognition systems may extract, from a 3D virtual model, CAD features and the related data used for the computational phase. The 3D CAD feature recognition procedure is described in detail in the following Section 2.2. It is worth mentioning that the current study refers to "assembly feature" to be recognized (i.e., the relative position of plates/component) and related parameters (i.e., the thickness of first plate/component vs. thickness of second plate/component) that can be read by the analysis of a 3D model (B-rep representation).
In addition to these clusters, another item of the classification (label) characterizes the rule-the rule type. This attribute may assume the following values: (i) information, (ii) warning, and (iii) critical. "Information" is a type of recommendation that would be desirable. However, it does not affect the processing of the component, nor its cost. A "warning" is an important aspect to address since it saves wasting manufacturing time and cost. Still, it does not prevent component manufacturability. "Critical" is the most significant rule, actually stopping the component manufacturability.
An example may facilitate the understanding of the type of information to insert for each section. The DFW rule reported in the table is the W025, and it is defined as the following: "Avoid weld beads crossing." For the first cluster of information, in this specific case, the rule is valid for all the arc welding processes (technology type-level I), independently from the specific type process (technology type-level II). For the second cluster of information, again the rule is applicable in general for all materials (material class) independently from the specific type (material type). For the third cluster of information, CAD features to recognize from the CAD models are (i) the weld beads and (ii) the position of the weld beads. No information is required regarding the PMI, while for parameters, it is required to assess the distance between the weld beads (D) in all the directions (x, y, z) and the dimension of the weld beads (if more dimensions are available, the largest one is recorded). Thus, the rule that discerns validated and non-validated rules is D ≥ 2,5 Z. When D is less than 2,5 Z the rule is non-validated (the weld beads are crossing other beads of the heat-affected zone); otherwise, the rule is always validated.

Knowledge Representation Phase
The knowledge representation phase begins with the definition of a predefined form for each DFW rule. It refers to the description of knowledge intended to be understandable for designers and engineers and then facilitates the implementation of corrective design actions. This phase encompasses the logical definition of DFW guidelines (syntax) and related information (e.g., suggestions about design changes to guarantee product manufacturability). A standard form, one for each design rule, is defined according to a given taxonomy and syntax, which are necessary to keep consistency among different guidelines. In this manner, the rules contain the same level of detail and information that can be manipulated by the mechanical designer during the product development process. DFW guidelines syntax requires mandatory and optional information. Mandatory information provides the minimum set of data to perform a design improvement. The data are (i) the design action to perform (verb) and (ii) the subject which requires modification (name). Optional information provides additional data that allow for the clarification of the context in which the design action is needed. These data include (i) the manufacturing process, (ii) the type of feature involved, (iii) the type/family of part, and (iv) the type of material.
To give a detailed understanding of the knowledge representation phase, Figure 3 presents the DFW guideline syntax and an explanatory picture illustrating what to do and what to avoid. The same example is reported here concerning the W025 DFW rule: "Avoid weld beads cross in arc welding."

CAD Feature Recognition Method
Feature recognition is the procedure used to extract features from a geometric model. As illustrated in Figure 2, through this step, it is possible to retrieve information from a 3D CAD model and to connect product features with the DFW rules. A feature recognition procedure begins by defining the types of features to be identified. Currently, a shared methodology for feature classification is not yet available because it depends on the application scenario [35]. In the past, researchers have defined multiple categories of aspects that were used in this paper to realize a feature recognition framework. Many studies were focused on the feature extraction for machining operations [36][37][38]. However, feature recognition for welded products has had its share of interest within the academic world [39][40][41]. Figure 4 illustrates the types of features used in this research and their relationships. Before proceeding with the description of the method, it is worth noting that the application of the feature recognition system is dependent on the way in which the 3D CAD model is developed. Indeed, two scenarios are possible when a welded product is modeled using a 3D CAD system. The first scenario, which is the one considered in this study, refers to a 3D model where the weld bead (or the weld point for resistance welding) and the manufacturing information are included (see Figure 5). Indeed, commercial CAD packages include functionalities to input welds and represent them through solid volumes or annotations in 3D representations of products. The user usually selects a pair of faces, and the system proposes alternative weld bead shapes to join the faces. The beads are created as associative features for drawings and are taken into account in computations of mass properties of products. The second scenario refers to a 3D model where the weld bead is not modeled and all the necessary information for the manufacturing phase is included only in the 2D drawing issued for manufacturing purposes. This second scenario is still a standard practice in many design departments, even though the newly developed computer-aided packages provide a dedicated environment for the modeling of the welded product and the possibility to include welding PMI within the 3D model. Moreover, the fact that complex welded products require a collaborative design approach with the involvement of several operators and often the file exchange through a neutral file from and to different CAD tools makes this practice more widespread. Indeed, in most cases, neutral files (i.e., stp, iges, or stl) do not allow storage of PMI, and all the welding data reported in the 3D model is lost when converting. However, this second scenario is not considered in this research due to the limits in the analysis of the welding features and parameters. It is worth noting that some researchers have tried to rebuild welding features from a neutral file (i.e., STP) even if some important information needs to be manually inputted by the user [41]. Based on this limitation, only new product development processes or design reviews that start from native files with the PMI related to the welding process are considered.
The first type of feature to recognize belongs to the cluster "component feature" and refers to each item (part) of the assembly. A component feature is a physical attribute used to represent the component [42]. It describes the most relevant characteristics of a component, such as its material (i.e., material feature), mass, volume, and area. Each part of the welded assembly has only one component feature (i.e., only one mass, only one volume, only one material). Component features can be extracted from a 3D geometry (boundary representation (B-rep) model) because attributes included in this class are readily available.
The second type of feature to recognize belongs to the cluster "geometric feature" and refers to each item (part) of the assembly. Geometric feature refers to a specific kind of form feature (i.e., a feature that embodies elements characterized via shape properties) used for representing information related to a given geometry (e.g., hole, pocket, slot, thread) [32]. Each part of the welded assembly has one or several geometric features and they are characterized by the type, the list of faces, the list of properties, etc. Geometric features can be extracted from a 3D model by using specific software tools for geometric feature recognition (e.g., SolidWorks by Dassault Systems has a module for feature recognition).
The third type of feature to recognize belongs to the cluster "welding feature" and refers to the product. welding feature describes a specific kind of form feature that is functional to assemble different components (e.g., the weld beam) [35]. It describes a specific welding feature through its type, list of properties, etc. There is at least one feature for each assembly that joins two or multiple components. As previously mentioned, welding features can be extracted from a 3D model by analyzing the manufacturing properties (PMI) directly inserted by the designer through the software.
The use of an application programming interface (API) allows the reading of CAD data from the CAD application and the linking with the dedicated DFW tool. The DFW tool will be able to compare product features and related parameters with the rules stored in the DFW database with the aim to verify if the product design is compliant with the welding process.

Case Studies and Results
The adoption of the proposed approach allows the definition of 46 DFW rules, all in the class of fusion welding. The overall list of DFW rules is reported in Appendix A. The structure of the DFW rules list (Appendix A) is the same reported in Table 1, which presents the overall structure of the repository used for collecting and storing the rulerelated information. In the first column of the repository are stored the rule number, which is a positive, progressive number. The central columns contain three clusters of information, namely, (i) manufacturing technology, divided into manufacturing technology class and manufacturing process type; (ii) material, composed by material class and material type; and (iii) CAD feature recognition, identifying geometric parameters and thresholds associated to a given rule. In addition to these sections, information about rule type is stored (i.e., info, warning, critical). The last two columns contain the guideline and the reference. The first represents the suggested guidelines displayed to the designer, while the reference is the rule's source (reference).
It is worth noting that this rules list is only a preliminary study and more rules will be added in further development to consider industrial knowledge, which is a crucial and long-term activity in the definition of DFW rules. Considering the guideline sources can be noticed that the majority (33 of 46) came from handbooks and reports , 10 rules came from company practices and only three rules came from other sources (i.e., DFMPro tool) [47]. In this preliminary version, the distribution of the DFW rules among different fusion welding technologies is the following: (i) overall, 29 have general validity for all welding technologies, (ii) eight refers to resistance welding, and (iii) nine refers to arc welding. Among the nine DFW rules referred to as the arc welding processes, six belong to the gas metal arc welding (GMAW) process. On the other hand, among the eight DFW rules referred to the resistance welding processes, all of them are specific for the spot-welding process. Concerning the rules distribution among different materials can be noticed that the majority of rules can be applied to all metal materials, while only three rules are focused on a specific welded material (W031, W032, and W033). These rules define the gas mixture for a particular welded material (aluminum alloys, carbon steel alloys, and stainless-steel alloys).
The rules presented in Appendix A can be also classified in relation to the CAD features to recognize. Many of them use plate thickness (12 of 46) as the main feature to recognize and/or bead dimensions (7 of 46). Rules focused on bead dimensions' features could be also divided based on bead parameters. Most of them (five of seven) consider bead height or thickness, four are focused on bead width and two rules use bead length as the parameter. It is worth noting that each rule could be considered more than one CAD feature or one parameter at the same time. Following the collection of these DFW rules through the knowledge-based approach previously described, the overall approach has been applied to a couple of real case studies.

Case Study 1: Jack Adapter of Heavy-Duty Prop
This first case study refers to a product (heavy-duty prop) composed of several welded sub-assemblies ( Figure 6). Heavy-duty props used in construction may consist of a girder beam body having a square section formed by four tubes at the corners held together by welded braces on the four sides. Usually, the total prop length can be changed by adding or eliminating extensions. At either end, the prop presents an adapter that concentrates the axial force on the four corner tubes onto another device, which can be a base plate, an adjustment jack, a hydraulic jack, etc. In this case study, a jack adapter is considered, i.e., an adapter capable of receiving the four corner columns and transmitting their axial load to a single jack, or threaded tube, acting on the adapter center. This solution is achieved by designing the adapter as a hollow box comprising two main faces, top and bottom, four tubes in the corner, apt for receiving the girder corner columns, and wall plates and strengthening gussets for holding them together and conferring to it the desired strength. Needs related mainly to on-site handling and costs, require the adapter to be as light as possible; hence, its design is based on thin plates and critical welding.
In this study, only one welded sub-assembly has been analyzed in detail, with the aim to provide a global overview of the application of the methodology. The analyzed sub-assembly, called jack adapter, is the welded assembly at the base or at the top of the heavy-duty prop (see (Figure 6; framed in red). This item presents typical issues that can be highlighted in the early design phase of the product development process. The jack adapter is composed of 11 components made with the same material (carbon steel low grade). Some of them are commercial components (i.e., the center tube and the half sleeve), while others are machined with a laser cutting process. The CAD model and the bill of material of the jack adapter are presented in Figure 7. The welding process used to manufacture this assembly is the gas metal arc welding (GMAW) with manual operations.  Starting with the feature recognition process, the following features were retrieved by the analysis of the welded product: (i) component features, (ii) geometric features, and (iii) welding features. The component features (Table 2) are related to the 11 parts which compose the assembly and only six parts are unique, as shown in Figure 7. Each component is made of the same material (carbon Steel S235JR) with a similar weight (from 0,39 kg to 1,73 kg).
For the sake of brevity, Table 3   The welds have different lengths and widths in the functions of connected components. Lengths vary from 135 mm to 263 mm, while welding widths vary from 3 mm to 4 mm according to the degree of mechanical strength required.
By following the proposed approach, after reading the data related to the CAD feature recognition, an analysis of the features against the DFW guidelines was carried out. For the assembly analysis, only the set of DFW rules referring to arc welding technology calls, including the specific rule set of GMAW technology type was selected. Then, thresholds of parameters that are characterizing each DFW rule are checked to the features identified in the feature recognition phase. Thus, explicit knowledge characterizing each DFW rule is checked with the feature identified in the feature recognition phase. Within the rule analysis phase, three design issues (non-validated DFW rules) were highlighted regarding the jack adapter.    The first issue is only the "info" rule type and refers to the use of specific shielding gas for the GMAW process. Indeed, this information is not connected with the geometry of the model but regards the materials and the process. Indeed, for the GMAW process of carbon steel, it is recommended to use CO 2 (100%) or a CO 2 /argon (25-75%) mixture as a shielding gas. The non-validated DFW rule reported here is the W033.
The second issue is a "critical" rule type and refers to the slot welding of the web plates (component C002) with the center tube (component C004). In this case, the dimension of the slot is non-compliant with the overall dimensions of the two components and the structural performance of the weld could be compromised. In particular, the issue lies in the slot width which is below the threshold defined by the rule. In the original design, the slot width is 7 mm and the thickness of the plate is 4 mm. Based on the rule, the minimum slot width for slot weld should be three times the thickness of the plate and so 12 mm. Another option to cope with this issue is the possibility to remove the slot weld and to fix the center tube (component C004) with the web plate (C002), which should be made with a different geometry. The new web plate (a straight plate without the bend) could be used to connect directly the half sleeves with the center tube. This is the W038 rule of the DFW rule DB reported in Appendix A. The new weld is then recognized through the feature recognition system as an assembly feature and the validation process of the DFW rules can be restarted. After the new run of the analysis, only the first two information rules are displayed by the method.

Case Study 2: Lateral Frame
This second case study refers to a product (a component of a one-sided bracket) composed of several welded sub-assemblies. Sometimes, the construction of certain infrastructures requires pouring concrete into a one-sided formwork or mold without the possibility of contrasting the huge pour pressures by use of tie rods, anchors, etc. This is, for example, often the case in building cut-and-cover tunnels underground. The concrete is poured between the excavated earth (on the outermost side of the tunnel) and the formwork, which delimits the inner side of the tunnel walls. In those cases, one-sided brackets are often used, consisting of a trapezoidal girder bracket comprising front and back beams connected by braces, as shown in Figure 8. Should such a bracket collapse, the consequences would be catastrophic in terms of safety, project cost, project delay. As a consequence, these brackets, albeit rather simple in terms of design and manufacturing requirements, present very critical requirements to eliminate the possibility of weld failure. Analyzed in this case study is the back component of this type of one-sided frame, which itself is made of a standard section I beam with various stiffening gussets and other components welded onto it.
The analyzed sub-assembly, called a lateral frame, is the welded assembly sloped at the external part of the bracket (see Figure 8; framed in blue). This item is representative of different issues that can be highlighted in the early design phase of the product development process. The lateral frame is composed of 12 components (only six are unique) made with different materials. Components 1, 2, 3, and 4 are made in carbon steel S235JR since they require higher mechanical strength due to their structural role, while components 5 and 6, which are used only for structure handling, are in aluminum alloy 6061. The main body of the assembly is a commercial component (e.g., HE 180A) while others are laser cut plates. The CAD model and the bill of material of the lateral frame is presented in Figure 9. The welding process used to manufacture this assembly is the manual gas tungsten arc welding (GTAW). Starting with the feature recognition process, the following features were retrieved by the analysis of the welded product: (i) component features and (ii) welding features. For the sake of brevity, the geometric features table was not reported within the manuscript but retrieved as well. Table 5 analyzes the component features of the lateral frame. In this case study, part weights vary in a wide range, from 0,10 kg (component C005) to 152,01 kg (component C001), which shows high variability in mass.     Again, as for the previous case study, the welds have different lengths and widths in the functions of connected components. Lengths vary from 92 mm (W004) to 648 mm (welds 3-8 of W001), while welding widths vary according to the degree of mechanical strength required-W001 welds are 8 mm and 7 mm wide, W002 and W003 welds are 4 mm wide, while W004 welds are 6 mm wide.
As was performed for the previous case study, by following the proposed approach, after reading the data related to the CAD feature recognition, an analysis of the features against the DFW guidelines was carried out. For the assembly analysis, only the set of DFW rules referring to arc welding technology calls, including the specific rule set of GTAW technology type was selected. Then, thresholds of parameters that are characterizing each DFW rule are checked to the features identified in the feature recognition phase. Thus, explicit knowledge characterizing each DFW rule is checked with the feature identified in the feature recognition phase. Within the rule analysis phase, three design issues (nonvalidated DFW rules) were highlighted regarding the lateral frame.
The non-validated rules identified within this welded assembly are different than the previous example.
The first rule (W041) is a "critical" rule type and it refers to the use of different materials that need to be welded. Indeed, the material used for the C001 component (the HE-180-A section bar) is carbon steel, while the material used for component C005 (crane eye) is an aluminum alloy. This could be a mistake from the technical department in the selection of commercial components from the company repository. Indeed, it is well known that dissimilar materials cannot be welded. However, the use of this system is able to catch also mistakes that come from repetitive, routine work.
The second non-validated rule (W043) is a "warning" rule type and it refers to possible distortions of the final assembly when components with different thicknesses are welded. In the original design, the component C001 (the HE-180-A section bar) and the component C004 (plate) show different thicknesses, respectively 9.5 mm vs. 6 mm. By upgrading the thickness of the less expensive part (i.e., the plate) to 9.5 mm, it is possible to cope with this welding issue and avoid possible deformation of the structure during and after the welding process. The weld is now called "Weld beam 03_ HE 180 A vs. Plate 3_mod" after modification is reported in Table 7, and the validation process of the DFW rules can be restarted. After the new run of the analysis, no non-validated rule is displayed by the method.

Discussion
In today's globally competitive market, the development of more challenging and complex products is pushing designers and engineers to shorten the design phase and to close the gap between design and manufacturing departments. To do so, computer-aided engineering and design tools are more and more important as they allow us to anticipate post-design concerns and manufacturing issues. Manufacturing knowledge is one of the most important assets of a company and the way to collect it and make it available for designers and engineers is a key topic. Thus, the proposed rule-based system for the design of welded structure aims to capture, retrieve and suggest design rules according to the given context of the welded assembly. The DFW method and tool proposition allows making accessible manufacturability knowledge within the design phase (3D modeling). The proposed approach enables the identification of design issues, analysis of the DFW rules propagation in the CAD design environment, and distribution of "know-how" to designers in the context of their specific design activity (explicit knowledge). This activity is on the critical path of the engineering design process; thus, the developed method is supporting knowledge-intensive and prone to error engineering tasks, extending the current CAD capabilities. The adoption of the proposed approach highlights several outcomes. The first is related to the effort and time required for developing welded compliant products. With this approach, design review loops may be reduced, thus improving the product timeto-market by diminishing knowledge-intensive engineering practices. The advantages include the integration of concurrent engineering concepts in welding design and the possibility to improve the overall efficiency of design and manufacturing practices. The method developed in this study shows great potential for real-world reduction of product cost and lead time through the implementation of concurrent engineering and design for welding concepts in the early design stages in a complex welding environment. Another interesting outcome concerns the possibility to share manufacturing knowledge across members of a design team and reuse it each time it needs.
Two case studies are proposed within this work to demonstrate the benefit introduced by the application of the proposed approach to real products made of few components.
The two examples show how concerns can be detected by the adoption of the proposed approach and how design review steps can be skipped. However, for the time being, few limitations are observed. The first limitation deals with the fact that the method was applied to the welded assembly made of few components and simple geometries. Indeed, the feature recognition system works well for simple shapes and may fail for complex shapes (i.e., new feature types and additional properties to consider). This is strictly related to the new DFW rules that can be defined that may require the recognition of different feature types or interactions between features. The second limitation lays in the scarce availability of DFW rules. Indeed, in this study, only 46 rules were retrieved and most of them were retrieved by the analysis of the literature. The most relevant action to this aim is the collection of implicit tacit knowledge within the company by the use of dedicated activities such as interviews with key employers, or the analysis of the most challenging projects developed in the past. Due to these limitations, future studies will focus on three main topics, namely, (i) enlarge DFW rules collection, (ii) extend the feature recognition system through the identification of new features, and (iii) software implementation and development.
In particular, concerning the last point, the approach presented in this paper is the backbone of a software tool for virtual assisting designers in evaluating possible design inconsistencies with welding processes.

Conclusions
The papers describe a method and a tool proposition that enables the identification of welding design issues during the CAD development of a welded product/structure. This system allows the analysis of the rules propagation in the CAD design environment and the distribution of "know-how" to designers in the context of their specific design activity (explicit knowledge). This research encompasses several disciplines (e.g., welding knowledge classification, 3D model feature recognition, computational analysis) and proposes a framework to develop an engineering design tool extending the current average CAD capabilities. The adoption of the proposed approach highlights several findings. The first is related to the effort and time required to develop welding-compliant products and avoid welding operation issues in the construction yard. With this approach, design review loops may be reduced, thus improving the product time-to-market and the overall product quality. Another interesting outcome concerns the possibility to share manufacturing knowledge on welding technology across members of a design team and reuse it each time it is needed. With regards to this advantage, there is the possibility to use the proposed approach for teaching initiatives and to educate the young generation of designers with a learning-by-doing system. The case studies presented in this paper refer to two metal structures (a jack adapter of a heavy-duty prop and a lateral frame of a bracket structure) fabricated with arc welding processes. The increasing complexity highlighted for managing these products points researchers toward the adoption of algorithms and software tools for automating the methodology presented in this paper (e.g., software integrated with 3D CAD system). The approach presented in this paper is the backbone of a software tool for virtual assisting designers in evaluating possible design inconsistencies with manufacturing processes. Few limitations were observed with the development of the proposed methodology and framework. The first concerns the daily update of the DFW rules repository. This process requires the continuous analysis of new documents to retrieve additional tacit knowledge that can be translated into explicit knowledge by the use of the proposed knowledge-based system (rule insertion form). This drawback requires the involvement of a dedicated person who is in charge of new rules collection and database updates. The second one concerns the possibility to estimate the saving gained by the implementation of the proposed guidelines. Indeed, the estimation of economic key performance indicators (KPIs), such as the manufacturing cost starting from the analysis of 3D product features, will allow quantifying the differences between the original design and optimized design. Due to the limitations mentioned above, future studies will focus on two main topics, namely, (i) enlarge the DFW rules collection and (ii) assess KPIs based on design parameters (i.e., costs).  Avoid chamfer where not required in machining process [46]