Process selection methodology for near net shape manufacturing

The generic process selection procedure usually has three steps: screening, ranking and a search for supporting information [6, 27]. Seven general approaches have been identified for selection of the best process for a given material, design characteristics and product requirements: Considering each of these in turn, the analytical papers develop a multi-variable system of equations, quantifying the different process features and their capabilities of matching with the component requirements. This formulation allows the process with the resultant lowest cost to be considered as the best candidate.

For example, [2] develop a model based on manufacturing cost prediction. The model provides the material (i.e. considering only the product volume and the material cost) and basic processing costs, depending on the selected processes and cost is calculated through the cost time and production volume, using empirical constant. This cost refers to the production of an ideal component design for the selected process and a coefficient (relative cost coefficient) corrects the process cost considering the geometry to produce. The coefficient is composed of four parameters (determined though empirical graphs), associated with geometrical shape, section reduction/thickness, tolerances and surface finish giving the distance between the current and the ideal conditions. Swift and Booker [70] use the [2] formula, introducing a matrix for a preliminary screen of the processes. Other authors use cost functions and cost estimation for pre-selection screening: for example, [40] provide a measurement of casting process compatibility for the required production volume, weight input, thick/thin sections, tolerances, and surface finish. A simple proportion between the available capabilities and the requirements give a compatibility score. Every characteristic is weighted, depending on its importance, with a qualitative system. The casting processes are thus ranked depending on their compatibility values. Rao and Padmanabhan [58] use graph theory and matrix approach for screening the additive layer manufacturing processes. The combination of these two methods is able to deliver a multi-criteria decision, defining the interactions between selection attributes.

The process attributes can be either qualitative or quantitative and their responses to the product requirements as well as their interrelation are summarized in an single index. For ranking the casting processes, [27] use the cost function with compatibility ranges for identifying the possible feasible processes and ranking them. From these first step, the best process is selected and, in a subsequent step, the process technological (tolerances, workable dimensions, surface roughness) and economical capabilities are matched, giving a complete overview of the process ability for produced the required product. The work is based on a previous investigation by [26], comparing target design and product requirements (material, size, product shape, mechanical precision and cost) with those that lie within the capacity of a large number of processes, seeking the subset which is capable of making the component. The subset is then ranked by economic criteria.

The probabilistic approach aims to develop a statistical correlation between the process capabilities and product requirements. In particular, fuzzy logic (artificial intelligence technology in control systems and pattern recognition) had been used by several authors. It is based on the observation that people make decisions based on imprecise and numerical information [19]. Fuzzy models, or sets, are mathematical means of representing vagueness and imprecise information, hence the term fuzzy [39]. Different from traditional probability, fuzzy sets are capable of representing, using and manipulating data that has a range of values, due to their uncertainness. Hence, in fuzzy logic, distinction between from full compatibility (one) and incompatibility (zero) is gradual between extreme ranges of the fuzzy set. Figure 1 illustrates the fuzzy logic approach. Several authors applied slightly different versions of fuzzy approach to process selection and decision making in manufacturing [19, 30, 54, 60, 73]. Where, L₍min − abs) is the absolute minimum value, L_mi n is the minimum typical, L_max is the typical maximum value, and L₍max − abs) is the absolute maximum value of the investigated process’s feature. L_req is the requested value of product feature (e.g. required surface roughness). Compatibility assessment can be performed by mapping from qualitative description (‘low’, ‘low to medium’, ‘medium’, ‘medium to high’ and ‘high’) to numerical values.

Giachetti [30], Ravi [59] and Daws et al. [19] define compatibility by the requested value and four values which define the fuzzy set. If the requested value is outside of the set (4), compatibility is considered null. If it is in normal range, then the request is fully compatible (1). If value falls between normal and extreme ranges, then the value is intermediate between 0 and 1, defined by a linear behaviour (2, 3). Using [22] possibility theory, [30] defines two different cases that occur in compatibility evaluation: possibility and necessity are defined. Possibility assesses to what extent a feature satisfies the request (optimistic selection strategy); on the other hand, necessity expresses to what extent a features certainly satisfies the query. The latter adapts pessimistic selection strategy by measuring the impossibility of the opposite event, determined using the complementary probability of the event itself.

Figure 2 illustrates the process of performing the calculations, using Eqs. 1 to 4 to determine the possibility and necessity values for a linear request. In order to evaluate values of possibility and necessity, a unique compatibility number is required, [30] used a factor called β that represents the level of optimisms or pessimism that is acceptable to the decision maker. Factor β is 1 for an optimist decision maker and 0 for a negative one (so always a value in the interval β ∈ (0,1)). A weighted average is calculated for each requirement between possibility and necessity values, mediated by factor β (possibility) and 1 − β (necessity). Using this methodology, a compatibility measure has been assigned to every process/product selection features. A geometric weighted mean is used for aggregating all n th compatibility values (5). Weight (w) is assigned to every feature using linguistic values. Each of them is calculated as in Eq. 6. Giachetti [30] applies this theory to the first stages of product design and process selection, including a broad range of processes and material as possible candidates. Perzyk and Meftah [54] use fuzzy logic for developing design for manufacturability of a single component. Functional requirements, manufacturing rules and material processability are evaluated for a single component through a process index, taking into consideration evaluating production volume, appearance, surface properties, dimensional tolerances and material structure. The index is a triplet-type fuzzy number, which is combined with the ideal process (depending on the product requirements). Daws et al. [19] limit the search to casting processes, including investment, mould (permanent, ceramic and full), shell, sand, die and squeeze casting. Similarly, [60] apply a fuzzy logic approach to the cutting process selection that considers the material-thickness relation, cutting speed, piece complexity and process tolerance capabilities.

Knowledge-based systems use empirical data (usually collected in databases) in order to support selection process. Knowledge-based systems are usually flexible and leave the decision making process to the user, providing all the required information to operate. A pioneering knowledge-based system in this area was reported by [65] to aid designers in choosing the best alloy and casting process for a particular set of specifications. The database displays both numerical and linguistic description of the processes, suitable for a certain material. The database includes a list of available material, selecting priorly to the manufacturing process (i.e. first list of processes are the material compatible ones). The designer selects qualitatively the best processes from its description, having excluded the unsuitable ones (i.e. relating to material and product specifications). Yu et al. [79] develop a computer-based routine which connects the geometrical factors, material and production factors attributes for identifying the most suitable process (i.e. selecting from casting, hot and cold forging processes). The algorithm uses a developed design compatibility analysis which quantifies the compatibility of every analysed category, by comparing the required values with datasets, for every considered process. Darwish and El-Tamimi [17] propose a knowledge-based algorithm for casting process selection, basing their decision criteria on design, production and manufacturing attributes. The author compare the process manufacturing attributes quantitatively (minimum thickness, tolerances, mass range, surface roughness, economic lot size), qualitatively (porosity, dimensional accuracy, mechanical properties) as well as the cost (tooling, labour, finishing and scrap costs). The available range of materials is used as a screening criteria for the processes. Similarly, [25] develop a system for casting process selection including a comparative cost routine (Fig. 3). According to the previous papers, the authors screen the processes on different levels (casting alloy, geometric complexity, casting accuracy, production quantity and comparative costs). In contrast to other researchers, geometric complexity has been quantified through questions regarding the product (e.g. undercuts or internal holes presence). The selected material has been used as screening factor, taking into consideration the resulting and required mechanical properties. Xu et al. [78] develop a knowledge-based system for additive layer manufacturing, including the process cost as a decision criterion.

Complexity measurement is another tool adopted by researcher for quantifying and ranking manufacturing process in order to select the most suitable. The complexity approach logic defines the lowest process chain complexity (including process and product design) as easiest from manufacturing the component. Product complexity influences directly the manufacturing complexity, so an effective understanding of complexity nature and its relative measure can directly connect them. Product complexity increases with the number and diversity of “features” to be manufactured, as well as the nature and difficulty of the tasks required to produce the features [24]. Cooper et al. [16] have measured product complexity as a volume weighted average; meanwhile, [32] has used entropy for the information number evaluation. ElMaraghy and Urbanic [24] developed a complete formula for evaluating the product complexity by measuring its entropy (given by the information number and uniqueness o features) and the complexity of each of its features. Features and specification are defined and evaluated for every characteristic, assigning them a factor (0 low effort, 0.5 medium effort, 1 high effort). All the factors are incorporated in the feature complexity coefficient and weighted by their percentage of presence in the component.

A matrix methodology is used to determine the relative complexity coefficient [24]. Complexity matrix describes all product characteristics and specifications. A factor indicates the relative effort to produce each of them or to perform the related task. Features and specification are defined and evaluated for every characteristic, assigning them a factor (0 low effort, 0.5 medium effort, 1 high effort). All the factors are incorporated in the feature complexity coefficient and weighted by their percentage of presence in the component.

The complexity index (obtained through the correspondent matrix) represents the difficulty of producing the component. A complexity index number does not have any meaning by itself. Comparing processes’ complex indexes defines the closest one to the final shape, in terms of less needed manufacturing effort. Wiendahl and Scholtissek [75] expand the complexity concept to the whole manufacturing process, including product design, operation (process equipment, tools and labour) and structure. Similar to the previous authors, [48] quantity the manufacturing complexity using an entropic approach. Their model evaluates both the various component types and technologies used in a manufacturing system on the system’s structural complexity. The authors apply their model by selecting the lowest complex manufacturing system configuration of an engine cylinder head. Kerbrat et al. [41] use the manufacturing complexity for evaluating how to combine subtracting and additive layer process for producing a mould. A modular CAD tool has been developed for comparing every single feature of the mould, selecting the less complex one to produce. Guenov [32] identifies two measuring systems for high-level decision makers. The aim is to compare alternatives during pre-competitive studies or during the architectural design process of composite systems. Similarly, the previous authors, the first measure is a complexity estimation of the Boltzmann’s entropy, meanwhile the second measure is intended to estimate the costs and benefits related to system’s performance.

Methodological investigations use a qualitative approach to determine the best process selection. The outputs of these papers consist usually in framework or flowcharts. For example, [1] develop a complex framework for material and process selection, taking into consideration the whole product life-cycle. The framework analyses the product life-cycle, diving it into three main phases: manufacturing, service and design/development. A dedicated part of the framework tries to rationalise the activities of requirements definition (design) and satisfaction (process). Xu et al. [77] develop a system for estimating the impact of different applications of rapid prototyping processes. Using product requirements and process cost, the methodology is able to quantify the process characteristics and compare different processes. Shercliff and Lovatt [64] define the interaction of process, material and design is peculiar to every category of processes (e.g. differences between casting and welding of an aluminium alloy). For the authors, the product requirements need to be matched one-by-one with the process attributes: these requirements can be design-related (e.g. mechanical properties or dimensional characteristics), production-related (e.g. production volume and production rate) or processing-related. Similarly, the attributes can be process-, material- or design-related. A pair matching is evaluated on technical feasibility, avoiding in-process defects, product performance (i.e. final product characteristics) and economic bases. Different from all the other authors, [44] try to develop a connection between process modelling and process selection. They define the cost models and technical models mostly used in process selection, in order to validate the process candidate. Chakraborty and Dey [13] use the Quality Function Development (QFD) chart, usually called house of quality, for matching the technical and design requirements as well as connecting them with the customer requirements. The authors developed a total score from this well-known quality enhancement instrument, using a score matrix.

Some authors have been able to implement process selection into optimization algorithms developing complex models for assessing the process applicability. Working on reconfigurable manufacturing systems, [9] use genetic algorithms and a simulation-based optimization for process planning for a single product type, taking into consideration market demand fluctuation and minimum production volume (i.e. for making the production feasible). The functions to optimize have been defined as machine usage and change costs, configuration change cost, tool usage and change costs. A genetic algorithm (i.e. optimizing product design and machines data) and demand simulation software (i.e. providing) provide the most economic chain configuration. Vinodh et al. [74] apply a fuzzy analytic network, using different criteria for evaluating the best process and the best supplier to select. Qualitative scores have been assigned to different criteria for evaluating the process/supplier. A matrix assigns the value to the process for every criterion and the algorithm rank the different possible combination. The selected criteria are coefficients that belong to business improvement, product quality, supplier service and risk.

Topological models describe how elements (Finite Element Analysis) are bounded and connected. These numerical analyses are used for describing numerically the product features (e.g. using rules of proximity, the FEM elements identify an undercut). In this way, algorithm can assess all features of a component and assess the best process for realizing them. Holland et al. [36] develop a CAD-based algorithm that can identify cost-effective manufacturing options for metal forming. The matching is based on database shapes and features. The optimal and economic processes to associate with every feature are stored in database as well. The orientation of the feature is determined by the algorithm. This determines the most suitable process, defining the forming direction and the realizable features. A similar approach has been used by [43], developing a process-oriented forming features in cold extrusion to develop a process selection module (CAPP). The module is able to detect feature shape, main dimensions, and volumes, connecting them with the best suitable cold extrusion process option (i.e. giving also an indication of the stage numbers).

Material selection investigations can be taxonomized using the same categories of the process selection. Regarding process selection, the most part of the author use the workable material as screening for the available processes. However, some of the authors include material selection in their process selection method: [1] include a material selection in its methodology approach. Giachetti [30] use fuzzy logic also for material selection, using a variable request for predicting the different final properties. This allows the author to extend their probabilistic approach to the material selection. Brechet et al. [11] review the material selection methodology, pointing out the efficacy of the multi-objective criteria selection. Ashby has pioneered this field with several works (some of them extended to material selection). Ashby [6] identify firstly some material performance index for materials. The author develop instruments for material selection, mapping the Young modulus on the density of different materials and the linear expansion on the thermal conductivity. The mapping is dependent on the final product requirements (thermal distortion): dedicated procedures need to be developed in order to measure the material attributes for the particular product design. Ashby [5] applies single and multi-criteria optimization to material selection. The authors derive, from the objective function, some differential equations, using multiple input variables and boundary conditions as constraints. In the multi-criteria selection, the solution of the equations are trade-off Pareto surfaces. As single target, the author use the minimization of the component mass. In their applications, the author uses the multi-criteria for minimizing mass and cost (determining a Pareto trade-off solution) or using combined parameters to minimize (depending on the component requirements and functionality). For example, the author uses product of cost and density on the elastic limit (i.e. square root) and the density on the elastic limit (i.e. square root), forming another trade-off Pareto selection. Kutz [42] review some quantitative methods for material selection, pointing out the fundamental role of expert systems and numerical assistance (databases and knowledge base selection). Lately, the usage of stochastic and heuristic algorithms to material selection has rapidly increased. For example, [50] apply Analytic Network Process (ANP) to multi-criteria selection: the material characteristics taken under consideration are density, thermal attributes (operating temperature, conductivity), physical properties, fatigue and mechanical characteristic. The network is able to establish a ranking of different materials for single product requirements.

In conclusion, fuzzy logic is capable of ranking the candidate processes in order of their features’ compatibility with requested ones. Usually, these features include technological and other quantifiable requirements (e.g. tolerances, surface roughness), although it can be easily extended to every required feature (e.g. material usage, labour cost). The compatibility values are able to rank the processes and materials for the given requirements. Fuzzy logic is also able to quantify the compatibility qualitative features compatibility and deal with uncertainty.

Complexity approaches have similar potential, although their application appears to be oriented to product redesign and supply chain simplification. Similarly, topological optimization merges CAD and features identification, being currently used in many software packages. However, it fails to analyse complex problems, where uncertainty is present. Analytical models are less subjective and achieved highest precision in quantification of process/material compatibly, particularly when few features are considered. However, they are limited in dealing with uncertainty and complex connections between options. Further, analytical papers are limited to consider few selection criteria into their selections. Optimization papers overcome this problem by relying on probabilistic and analytical models, merging them with numerical capabilities and iteration.

Qualitative, methodological and knowledge-based approaches are most flexible and capable of dealing with complex interactions between material, design nd manufacturing process. However, the inability of quantifying feature’s compatibility and generally low levels of subjectivity limits them to the selection of relatively restricted categories of process and materials.