1 Introduction

In machining process, the appropriate selection of processing parameters plays a vital role in ensuring product quality, minimising machining costs, and increasing productivity. The role of optimisation has been drastically enhanced due to the emergence of high strength materials (such as titanium, Incoloy, Hastelloy, and superalloys) and intricate manufacturing designs (3D- and 4D-shaped components) requirements [1]. Machining these materials using the conventional approach may not be able to fulfill the customer specification because the component’s finishing consistency is restricted. Substantial attempts on the use of handbooks for conservative machining conditions and machining tool selections at the production planning level are still progressing. Manufacturing industries have traditionally relied on shop floor machine tool operators’ expertise and abilities in selecting the optimum range of machining conditions and equipment. This leads to low productivity due to the unoptimised machining capabilities. Many obstacles are also encountered in searching for practical solutions, such as low machine-tool power, torque, force limits, and parted surface finish. Thus, to attain maximum output from any machining process, the use of optimal parameters is necessary [2].

Machining processes are fundamentally classified into two categories, namely traditional and advanced machining processes. Conventional machining process includes turning, milling, grinding, drilling, boring, and shaping, while advanced machining process includes electrochemical machining (ECM) and its different variant, electro discharge machining (EDM), wire electrical discharge machining (WEDM), abrasive jet machining (AJM), water abrasive jet machining (AWJM), and ultrasonic machining (USM) [3]. Advanced machining process can be further derived into hybrid versions in many researches. Both machining processes operate according to their mechanism, which made them ideal for producing certain materials that can also be limited in their use. The processes also consist of several input process parameters that directly and indirectly affect the performance and surface characteristics of the machined components. The combination of input variable and multiple responses also create a need for the multi-objective optimisation. Therefore, some optimisation methods are necessary to avoid the unorthodox way of choosing input parameters [4,5,6].

To mitigate limitations due to the random selection of input parameters, one of the considered methods is the optimisation technique. Fundamentally, two types of optimisation technique are used in machining operations, namely conventional (design of experiment and mathematical iterative search) and advanced techniques (metaheuristic search and unique heuristic search for problem). This review aims to highlight the benefits of using evolutionary techniques in various conventional and advanced machining processes. Section 2 of this paper provides a general overview of machining and its operational parameters along with the majorly discussed responses by past researches. Section 3 provides the detail of multi-objective criteria in machining processes and the use of evolutionary algorithm (EA) in solving multi-objective optimisation problems. Section 4 discusses the performance of EA in finding the optimum solution for machining processes along with the author’s critiques. Lastly, Section 5 summarises the overall paper and provides a future perspective for further research in EA for machining.

2 General review

Machining is a process to produce a part or component to the desired dimensions and surface finish from a work piece by removing the undesired material in the form of chips using a cutting tool. Almost 90% of all engineering components undergo machining process during fabrication. It is imperative to design the components in such a way that can increase the efficiency, enhance the tool life, and reduce the overall cost in machining. To achieve the target, optimisation is applied to the machining process to improve the machining efficiency and quality of output. Table 1 presents the overview of the reviewed articles on machining parameters optimisation using conventional and evolutionary optimisation techniques, whereas Fig. 1 provides the summary of 70 research papers (2010–2020) reviewed in the present work. It has been observed in Fig. 1 that 60% of the optimisation processes were applied to the conventional machining (turning, milling, and drilling), while 40% of the optimisation processes were applied to advanced manufacturing processes. Out of the 60% of the optimisation processes, 29% of the processes were on turning, 18% on milling, 11% on drilling, and 2% on grinding. For advanced machining processes, 13% of the reviewed papers discussed optimisation applied to EDM, 4% on WEDM, 10% papers discussing the AWJM, 8% on ECM, 2% on 3D printing, and 1% each are based on ECDM, ECDD, and laser beam machining.

Table 1 Reviewed research papers from 2010 to 2020
Full size table
Fig. 1
figure 1

Distribution of machining processes conducted by research articles

Full size image

Figure 1 shows that the optimisation techniques were applied more on the conventional machining processes (or computer control types of conventional machining) compared to the advanced machining processes. This is because, the conventional machining process has less complexity or limited input variables [3]. Machining parameters in every machining process may be different depending on the type of operation and mechanism. In typical machining processes such as turning, milling, and drilling, the standard machining parameters that can be controlled are feed rate, depth of cut, and cutting speed. These parameters are set before conducting particular machining parameters, which is a crucial step in ensuring a good quality of machining outcome. The effect of setup parameters can be observed from the work piece after the machining based on surface roughness (SR), duration of machining process, tool wear, and material removal rate (MRR). In this section, the effect of optimisation on the machining parameters is discussed.

From the review, majority of the studies were performed on the alteration of SR of the work piece due to the variation in machining performance. SR is considered as an essential quality characteristic in determining machining performance. Lower values of SR on the workpiece will help in improving the corrosion resistance, fatigue strength, creep behaviour, surface friction, and stress concentration [1,2,3,4,5]. SR is also an indicator of product quality. According to Kant and Sangwan [15], achieving a predefined SR below a specific limit generally increases power consumption exponentially and decreases productivity. This fact enables various researchers [12, 15, 17] to utilise the capabilities of multi-objective optimisation in minimising SR and production cost. Other frequently optimised parameters are MRR, tool wear, cutting forces, machining time, and cutting temperature, as shown in Fig. 2.

Fig. 2
figure 2

Machining outcomes studied in past researches

Full size image

3 Application of evolutionary algorithm on machining optimisation

In machining optimisation, the selection of variable parameters is considered as a crucial and essential task because it helps in evaluating the efficiency of the machining process. The chosen machining parameters are usually based on human (machinist) judgement and experience. However, the chosen parameters may not have resulted in the optimum machining process. As an improvement, optimisation is needed to be applied in machining process to improve its efficiency. Optimisation tools and techniques can be divided into two major categories which are conventional and advanced methods as presented in Fig. 3. Most of the conventional optimisation techniques are based on gradient search. The conventional optimisation technique has the possibility to converge in local optimum depending on the degree of nonlinearity in the objective function and initial guess about the starting point. Thus, the conventional optimisation technique did not promise global optimum.

Fig. 3
figure 3

Type of optimisation tools and techniques

Full size image

In the recent years, researchers tend to apply advanced optimisation to machining processes. From Fig. 4, out of 70 research papers (2010–2020), only 37% of the papers applied conventional optimisation methods, while 27% of the papers used combined conventional and advanced methods, and 36% of the papers applied advanced optimisation methods. Compared to the conventional method, advance method especially metaheuristic search is based on EA that mimicked the biological mechanism as mutation, recombination, and natural selection to find an optimal design within specific constraints. EA like genetic algorithm (GA), particle swarm optimisation (PSO), artificial bee colony (ABC), simulated annealing (SA), cuckoo search algorithm (CSA), firefly algorithm (FA), and glow-worm swarm (GWS) are implemented in optimising machining parameters. Every EA has its own survival step to escape from local optimum and has bigger search area without neglecting neighbourhood solutions in finding global optimum solutions. In the preceding topic, three most frequent EA optimisations applied in machining problems are discussed.

Fig. 4
figure 4

Distribution of optimisation on machining processes

Full size image

3.1 Genetic algorithm optimisation

GA is one of the engineering optimisation techniques, a type of guided random search method. It is suitable for solving multi-objective optimisation problems to search different regions of the solution space. Thus, it is possible to find a global maximum or minimum set of solutions with more variables that can be optimised at one time. Solutions of GA are illustrated using the Pareto fronts. A Pareto optimal set is a set of solutions that are nondominated solutions frontier. With the Pareto optimum set, the corresponding objective function’s values in the objective space are called a Pareto front.

The idea of GA was founded in the 1960s and 1970s by Holland [76], which was adopted from natural genetics and evolution. Randomly combined solutions are evaluated and the information is exchanged between solutions to obtain the optimal solution. From the randomly created initial population, unlimited solutions are produced for the initial guess. By this method, the search space may be explored and avoids local maximum/minimum [77]. Each solution is evaluated by calculating the objective function values. Solutions with fitter chromosomes have higher chances to be selected for the next generation. This process is done through the probabilistic technique, in which an intermediate population with a higher representation of the strong species is generated. The intermediate population is allowed to perform cross-over marking and modified through mutation and the next population set is produced. This procedure is continual until the termination condition is reached [78]. The flowchart of GA is shown in Fig. 5.

Fig. 5
figure 5

Flowchart of genetic algorithm

Full size image

GA can be applied by single-objective or multi-objective functions. Single-objective function minimises SR for Ti-6Al-4V by optimising radial rake angle, speed, and feed rate in the end milling machining process [8]. The researchers were able to decrease the minimum SR by 27% and 26% compared to the experiment data and regression modelling, respectively. The minimum SR was also 50% better than that of the response surface methodology (RSM). Zain et al. [8] proved that the EA can improve the SR using the same range of machining parameters. Sangwan and Kant [11] also compared the optimisation method between GA and integrated RSM. The energy consumption was minimised during turning machining by optimising cutting speed, feed rate, and cutting depth. The results of GA and integrated RSM were compared with the real data from the experiment. Both GA and integrated RSM have comparative error with experiment data by about 5.1%. The finding highlights that, besides finding optimum solutions, GA also provides a realistic result as tabulated in the conventional optimisation method. Mahest et al. [32] also confirmed that GA optimisation can search the optimum solution in the search area by conducting a comparison study between RSM and GA. In this study, the composite rotatable design (CRD) model was developed by RSM and further optimised using GA to develop minimum SR. The GA recommended 0.25 μm as the best minimum predicted SR value. The validation test shows that the experimental values are in good agreement with the predicted values.

Besides RSM, GA has also been compared with other conventional optimisation techniques such as Taguchi orthogonal arrays. Selvam et al. [45] used the Taguchi method and GA to minimise SR in machining mild steel with coated carbide tools. The research focused on optimising the crucial machining parameters such as number of pass, depth of cut, spindle speed, and feed rate. Table 2 shows the optimised result utilising Taguchi technique and GA optimisation. The result indicates GA can search for minimum SR up to 0.8393 μm, while Taguchi technique was able to locate SR up to 0.933 μm.

Table 2 Optimised result comparison between Taguchi technique and GA [45]
Full size table

GA can also optimise more than one objective function in a single algorithm. Multi-objective GA (MOGA) will find trade-off between objective functions that need to be optimised (maximise/minimise) by optimising the machining parameters. This feature is crucial if the machining process needs to ensure another machining outcome will be affected by optimising only single outcomes. Usually, if the manufacture wants to optimise machining time, coarser surface finishes are produced. By implementing MOGA, all aspects in machining will be maintained at optimum condition. Wang et al. [17], Zhou et al. [24], Kumar. [39], Narayanan et al. [40], Dhavamani and Alwarsamy [41], Reddy et al. [42], Fountas and Vaxevanidis [43], and Manav and Satish Chinchanikar [50] also applied MOGA in their researches.

3.2 Particle swarm optimisation

Another type of evolutionary algorithm frequently applied in optimising machining parameters is PSO. PSO is an evolutionary computation technique developed by Kennedy and Eberhart [79] based on a bird flock’s social system. The objective of the algorithm is to graphically simulate the graceful but unpredictable choreography of a bird flock. Each bird (representing a particle) keeps track of its coordinates in the problem space associated with the best solution (fitness) it has achieved so far. PSO is initialised with a group of random particles (solutions). Then, the searching process is done by the particles (potential solutions) flown through the problem space by following the current optimum particles. Each particle keeps track of its coordinates in the problem space, which are associated with the best solution (fitness) it has achieved so far as shown in Fig. 6. The PSO concept consists of velocity and acceleration as the operator to find the best solution in the swarm.

Fig. 6
figure 6

Flowchart of particle swarm optimisation

Full size image

Manav et al. [50] optimised the machining process using PSO and studied the effects of PSO in optimising cutting parameters such as cutting speed, feed, and depth of cut to minimise SR, reduce cutting force, and maximise tool life. In this work, PSO was able to find the combination of cutting parameters to produce the best machining products. Also, it observed that for AISI 4340, 174 m/min of cutting speed, 0.15 mm/rev of feed, and 1.0 mm of depth of cut will provide best machining setup to achieve 3.91 μm as surface roughness by generating 413 N tangential forces, 148 N axial force, and 227 N radial force that are needed and also has 49 min of tool life.

Nain et al. [49], Gopalakrishnan and Dhas [52], and Rao and Venkaiah [54] applied PSO for the WEDM operation and worked on the multi-objective PSO (MOPSO) in their studies. Nain et al. [49] studied the effect of pulse-on time, pulse-off time, peak current, wire tension, spark gap voltage, and wire feed rate on cutting speed, dimensional deviation, and wire wear ratio in the WEDM. The study confirmed that pulse-on time is a significant variable among all the input variables. The findings were also supported by the Taguchi analysis along with the MOPSO.

Meanwhile, Gopalakrishnan and Dhas [52] conducted a study on WEDM and focused on improving MRR and minimising SR on AA7075-activated carbon composite using PSO and ANOVA. The optimisation results show that PSO is a useful technique for predicting the outcome from the experiment and optimisation method for WEDM. Rao and Venkaiah [54] also applied PSO in optimising MRR and SR. This research found that the results using the PSO were better than that of the RSM. Malghan et al. [55] also investigated PSO optimisation effects with RSM for face milling operation. Cutting force, SR, and power consumption were optimised by optimising the machining parameters such as feed rate, spindle speed, and cut depth. The results indicate that the PSO is effective in predicting the optimum values for desired machining response.

Table 3 shows the optimisation results from the previous research that compared PSO with conventional optimisation methods. All these four research articles used PSO and compared or validated the findings with the conventional optimisation method. These researches proved that PSO has better methodology to find the optimum solution than the conventional optimisation (Taguchi method and ANOVA). In addition, for the PSO optimisation, the researchers applied PSO to the WEDM which is considered as an advanced machining process. It shows that PSO can be applied to the conventional machining and effective for advanced machining processes.

Table 3 Comparisons of previous work between conventional optimisation methods and PSO
Full size table

3.3 Cuckoo Search Optimisation

The CSA optimisation is one of the latest EA optimisation methods. It is a global optimisation based on cuckoos’ behaviour proposed by Yang and Deb [80] in 2009. The idea of CSA was originated by the way of cuckoos lay their eggs in the host nests. Some cuckoos have evolved so that female parasitic cuckoos can imitate the colour and patterns of the eggs of a few chosen host species [81]. This reduces the probability of the eggs being abandoned and therefore increases their productivity. Generally, the algorithm for CSA is described in Fig. 7.

Fig. 7
figure 7

Flowchart of Cuckoo Search Optimisation

Full size image

Saravanan et al. [61] employed the CSA optimisation by optimising MRR and surface roughness for WEDM. By finding the optimum parameter values of cutting speed, pulse-on time, pulse-off time, current, wire feed, wire tension, servo voltage, and servo feed, the MRR and SR were improved. This study was conducted using three types of materials (D3 steel, a high-carbon and high-chromium tool steel) and three different types of wire material: plain brass, zinc coated and molybdenum coated. The objective function developed by using multiple linear regression models was used in the CSA optimisation. Figure 8 shows that the molybdenum wire outperformed the other two wires by producing higher MRR and lower SR in experiment and CSA optimisation, which proves the efficiency of the simple CSA algorithm in optimisation.

Fig. 8
figure 8

a MRR for three different wires based on experiment. b Maximising MRR with SR as constrain based on Cuckoo Search Optimisation [61]

Full size image

Mohamad et al. [59] applied the CSA optimisation to optimise the machining parameters in the deep hole drilling process. In this study, the regression technique was used for the modelling and CSA algorithm for the optimisation process. This study’s machining parameters are feed rate, spindle speed, depth of hole, and minimum quantity lubrication (MQL), and the machining performance that optimised SR. This study employed nest number, n = 25, mutation probability, ρa = 0.25, scale factor, β = 1.5, and step (s) = 1 for the CSA algorithm input. The optimisation process based on CSA indicates that the mathematical model using 2-Factor Interaction (2FI) produced the minimum value of SR compared to the experiment.

Meanwhile, Sathish [58] investigated the parametric optimisation in machining using RSM and a novel hybrid cee colony cuckoo search (BCCS) algorithm. The method was tested in WEDM to achieve maximum MRR and minimum SR by optimising machining parameters such as pulse on time, pulse off time, peak current, and servo voltage for Nimonic-263 alloys. Compared to Saravanan et al. [61] and Mohamad et al. [59], this research applied a hybrid algorithm by consolidating ABC and CSA to utilise better optimisation outcomes. The CSA is adapted in ABC when the onlooker bee step has not refined better alternatives, leaves from the specific choices, and makes up the arbitrary number of scout bee order. The optimisation results using BCCS were superior to those of the RSM. The deviation of BCCS predicted and experimental values for MRR was 2%. Meanwhile, the deviation of BCCS predicted and experimental values was 7%. This also indicates that the optimisation outcomes from BCCS are realistic and practically achievable.

Huang et al. [60] also adapted two optimisation hybrid methods and applied teaching–learning based outcome (TLBO) and CSA algorithm for parameter optimisation problems in machining processes that include abrasive water jet, grinding, and milling operations. In the proposed hybrid algorithm, the main idea is to combine good search ability of CSA and the fast convergence rate of TLBO. The proposed algorithm mainly involves two parts, so that the solutions abandoned in the CSA will perform Levy flight to generate new solutions. For other solutions, the researchers used TLBO to enhance the local search ability of CSA. Thus, the algorithm becomes more practical for a wider range of applications without losing the attractive features of the original CSA and TLBO. Experimental results show that the TLCS obtained better solution than those previously reported in the literature, revealing that the proposed TLCS is a very effective and robust approach to address parameter optimisation problems.

Parameters of AWJM were optimized using PSO, firefly algorithm (FA), ABC, SA, black hole (BH), cuckoo search algorithm (CSA), biogeography based (BBO), and nondominated sorting genetic algorithm (NSGA) by Shukla and Singh [68]. In this study, parameters such as transverse speed, standoff distance, and mass flow rate were considered to obtain these parameters’ effect on kerf top width and taper angle. The results show that BBO performed better than the other algorithms in terms of convergence. Based on Table 4, other algorithms have found an approximate optimum solution to each other. Table 4 shows that BBO has a maximum convergence rate compared to the other algorithms for the considered objectives.

Table 4 Results of single objective optimisation of AWJM process [68]
Full size table

4 Discussion on performance of evolutionary algorithm (EA) in finding optimised solution for machining process

As discussed in the Section 3, optimisation using EA shows an improved result compared to the conventional optimisation methods. As EA method is relatively new, researchers tend to validate the outcome against the conventional optimisation method. All 27% papers that conducted the comparison between conventional and advanced methods (Fig. 4) produced better or comparable results from EA compared to the conventional optimisation method. Based on that outcome, it can be concluded that EA optimisation method is a reliable method for optimising machining parameters to produce optimum machining outcome.

Figure 1 shows that EA was mostly used in the traditional machining processes. Compared to the conventional optimisation, EA optimisation can optimise more than one parameter and achieve the target more than the objective function. Based on Table 1, cutting speed, feed, feed rate, and depth of cut were optimised in the conventional machining process such as urning, milling, and drilling, which affected the SR, machining time, MRR, and tool wear. Meanwhile, every advanced machining process has different parameters. The effect of each parameter on the machining process performance is still unknown due to the lack of study. For WEDM, the parameters involved in the optimisation are voltage, pulse-off time, pulse-on time, and wire feed rate, while for AWJM, the parameters are standoff distance, feed rate, and abrasive flow rate. Even though the advanced machining involves many machining parameters, the optimisation study of each parameter’s effect to the machining process is still limited.

Comparison of these optimisation methods is frequently conducted between conventional optimisations and metaheuristic optimisation algorithms. It shows that the metaheuristic optimisation outperforms the conventional optimisation due to the wider search area with parameters to control the search. The evolutionary optimisation was employed by the researchers due to its the ability to find the optimisation results in the search area. The search area for conventional and advance optimisation is the same for optimum combinations of variables, and the optimum solutions are among the feasible solutions. Each EA has its parameters/operators to maintain the search area and avoid being trapped in the local maximum/minimum. Since the EA is based on probability techniques, the operator in the algorithm plays a crucial role in eliminating bad solutions and ensuring good solutions are maintain in the population. These operators are also to balance the local and global search either by exploration or exploitation. Table 5 shows the comparison between GA, PSO, and CSA in population size and controlled parameters. The operators used in the algorithm will determine whether the algorithm will find global solutions and the time taken to search for the solutions. Table 5 indicates that the GA required higher population compared to the PSO and CSA to achieve optimum solution. In GA, the initial population will be randomly created and go through the operators (for selection, crossover, and mutation). Then, the subsequent population will be created. Selection and crossover operators are responsible for evaluating and improving solutions in the population by doing small modification to the population, which is also known as the exploitation step.

Table 5 Comparison of evolutionary algorithm
Full size table

Meanwhile, mutation operators will change a fraction of the population to the random population to avoid being trapped in the local optima. Compared to GA, PSO and CSA have better and sophisticated algorithm to avoid being trapped in the local optima; hence, it is justifiable that PSO and CSA do not require large population size. However, a drawback of having a more complicated algorithm is that the converging time to find the solutions will be longer, which is also supported by [82].

Optimisation of machining is always an important procedure to improve the machining response such as SR, MRR, power consumption while minimising the machining variables like cutting speed, depth of cut, feed rate, and cutting depth. In the past 10 years, most works that employed EA optimisation are the conventional machining (milling, turning, and drilling,). The comparison of EA with conventional optimisation like Taguchi techniques, RSM, and ANOVA shows that EA can provide better optimum solutions. This study shows that conventional machining is still relevant and widely used in the industry today. Furthermore, the simplicity of the conventional machining due to the low number of variables involved in the process makes the operation easier to control and monitor. Without denying the advantages of advanced machining, Reddy et al. [64], Thakur and Singh [65], and Balaji et al. [66] implemented EA to advance machining process such as WEDM and AWJM. The advanced machining involves higher number of parameters than the conventional machining. Therefore, the effect of optimisation is more significant in the advance machining compared to the conventional machining. This shows the growth of interested researchers from advanced machining to implement EA in their optimisation works. As the development in computational science is progressing, it is expected that new EA with better performance will be created. Thus, the researchers from engineering fields, especially from the machining area, should take advantage of the optimum machining process.

5 Conclusion

The adopted optimisation approaches are generally the experimental or/and numerical methods with discrete variations in the parameters of interest to obtain the desired lowest SR, highest MRR, lowest production cost, and fastest production rate. Based on the review, several conclusions can be drawn and summarised as follows:

  • WEDM is the most frequent machining process that has been applied in the optimisation procedure of conventional machining such as turning, milling, and drilling, as well as advanced machining.

  • Minimisation of SR is the most common machining outcome used in optimising machining parameters.

  • Even though GA can be considered a pioneer in EA, GA is still preferable until today despite the emergence of numerous new EAs.

  • Evolutionary algorithm optimisation outcomes are superior compared to the conventional optimisation method.

  • Every EA has its unique parameters to obtain optimum outcomes and the setup for every problem is also unique depending on the problem itself. It cannot be applied as a general solution to other problems.