Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach

Journal of Cleaner Production xxx (2015) 1e12

Contents lists available at ScienceDirect

Journal of Cleaner Production journal homepage: www.elsevier.com/locate/jclepro

Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach A. Garg a, Jasmine Siu Lee Lam a, *, L. Gao b a

School of Civil and Environmental Engineering, Nanyang Technological University, 50 Nanyang Avenue, 639798, Singapore The State Key Laboratory of Digital Manufacturing Equipment and Technology, 1037 Luoyu Road, Huazhong University of Science and Technology, Wuhan 430074, China

b

a r t i c l e i n f o

a b s t r a c t

Article history: Received 5 March 2015 Received in revised form 10 June 2015 Accepted 10 June 2015 Available online xxx

From the perspective of energy conservation, the notion of modelling of energy consumption as a vital element of environmental sustainability in any manufacturing industry remains a current and important focus of study for climate change experts across the globe. Among the manufacturing operations, machining is widely performed. Extensive studies by peer researchers reveal that the focus was on modelling and optimizing the manufacturing aspects (e.g. surface roughness, tool wear rate, dimensional accuracy) of the machining operations by computational intelligence methods such as analysis of variance, grey relational analysis, Taguchi method, and artificial neural network. Alternatively, an evolutionary based multi-gene genetic programming approach can be applied but its effective functioning depends on the complexity measure chosen in its fitness function. This study proposes a new complexity-based multi-gene genetic programming approach based on orthogonal basis functions and compares its performance to that of the standardized multi-gene genetic programming in modelling of energy consumption of the milling process. The hidden relationships between the energy consumption and the input process parameters are unveiled by conducting sensitivity and parametric analysis. From these relationships, an optimum set of input settings can be obtained which will conserve greater amount of energy from these operations. It was found that the cutting speed has the highest impact on the milling process followed by feed rate and depth of cut. © 2015 Elsevier Ltd. All rights reserved.

Keywords: Environmental sustainability Energy conservation Energy consumption Machining Computational intelligence Milling process

1. Introduction The subject of energy conservation is being extensively studied by climate change experts. This is because one of the factors contributing to drastic climate change is the extensive amount of energy consumption from the various sources (household appliances such as air conditioner, industry, vehicles). Therefore, for conservation of energy, reducing energy consumption from these sources is a problem of national interest. Among these sources, the energy consumption from the industry contributes significantly. It is learnt that the industry sector accounts for about one-half of the world's total energy consumption which has been doubled over the past 60 years (Fang et al., 2011). This is because with the world witnessing growth in manufacturing industries for meeting the stringent demands of customers, the demand of energy for driving

* Corresponding author. Tel.: þ65 6790 5276. E-mail address: [email protected] (J.S.L. Lam).

the essential operations including machining processes (turning, drilling, grinding, and milling) has significantly shot up. This leads to an unfavourable environment due to the release of toxic gases in the atmosphere (De Soete et al., 2014; Egilmez et al., 2014a, Kreiger et al., 2013). Among these operations, machining is widely focused and performed in the manufacturing industries. Excessive use of energy used in driving the machining operations has adverse impacts on the environment. It is known that machining processes have an efficiency of below 30% (He et al., 2012) and 99% of the environmental impacts is from the electrical energy consumed by tools in processes such as milling and turning (Kant and Sangwan, 2014; Li et al., 2011). The consequences are soil, air and water pollution which may render environmental and social problems. From this perspective, saving of energy can result in higher environmental performance and productivity (Lam and Lai, 2014). Performing the manufacturing operations for higher economic prosperity with least environmental impact has led to a new manufacturing paradigm of environmental conscious manufacturing (ECM) or sustainability manufacturing or green

http://dx.doi.org/10.1016/j.jclepro.2015.06.043 0959-6526/© 2015 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

2

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

manufacturing (GM) (Bhushan, 2013; Balogun and Mativenga, 2013). Survey studies (Davim, 2011; Garg et al., 2013a; Chandrasekaran et al., 2010) conducted on the applications of modelling methods in machining operations reveal that the main attention was on reducing the implementation cost of the machining operations by optimizing the surface roughness (manufacturing aspects) of the machining operations (Aykut et al., 2007; Li et al., 2000). Among the machining operations, the modelling of milling process received less attention. To the best of authors' knowledge, comparatively, very little emphasis was paid in optimizing the environmental aspects (energy consumption and cutting forces) of no et al., 2012; the machining operations (Siddique, 2012; Tata Egilmez et al., 2014b). It is known that energy is consumed during the different stages of machining such as during the machining, post machining and in idle condition to drive motors and auxiliary components. Machine tool is designed based on the peak energy requirement, which is significantly higher than the non-peak energy requirement of the machine tools. This results in lower energy efficiency of machine tools. Optimization of energy component in machining operations can result in wide application of lower energy rated motors/auxiliary components and thus can prevent wastage of energy and improve the environmental impacts of the machining operations (Garg and Lam, 2015; Kant and Sangwan, 2014). The performance of any industry is accessed by the quality of products being manufactured, tool life and energy efficiency of the machining operations. It is known that the efficiency of machining operations is lower than 30%, with almost 99% of the environmental impacts are from energy consumption (Kant and Sangwan, 2014). Lowering the product quality or the tool life does result in a reduction of energy consumption. However, it imposes a greater risk to the rejection of products and thus a threat to its reputation and sustainability in market. Therefore, there is a need to find a balance between energy consumption and manufacturing aspects (product quality, tool life, etc.) by effectively determining the appropriate input process parameters of the machining operation (Kant and Sangwan, 2014). In context of optimization of machining operations, the formulation of models representing the functional relationship between the outputs (manufacturing and environmental aspects) and inputs (input process parameters) is vital. Due to the complexity of the process, it is difficult to formulate models based on partial knowledge about its physics behind the process (Armarego et al., 2000; Ahilan et al., 2012; Bhushan, 2013). Some of pez de Lacalle et al. the useful works in this aspect was done by Lo  pez de Lacalle et al. (2005, 2006) developed the (2005, 2006). Lo diagnostics tool for researchers to detect and solve two unexpected problems during milling of complex parts. A data acquisition system was set up to record the position of tool and cutting forces simultaneously by dynamometric plate so as to correlate the geometry of surface machined and the cutting forces in three axes (X, Y and Z). The first problem is by changing the feed rate continuously; the correlation between cutting forces and part geometry was observed. Second is detection of engagement of unexpected tool engagement coming from previous semi-finishing operations in machining of complex parts. Zulaika et al. (2011) also proposed an integrated approach for design of productive and light weight milling machines. In this approach, the interactions between the process and machine was measured by a stability model which resulted in identification of mechanical design parameters that limit the productivity and must be met to target the desire productivity. The approach when applied to the re-design of the actual milling machine resulted in above 100% productivity and 13% less energy consumption due to mass reduction of above 20%. Similarly, an integrated framework for achieving the optimization of cost and

energy consumption of manufacturing systems was proposed by Tolio et al. (2013). Also, authors have studied the literature review on modelling the manufacturing and environmental aspects of the machining operations by the statistical and optimization based methods. It was learnt that the most widely used method is RSM because it can be applied on the limited set of experiments (Yan and Li, 2013; Campatelli et al., 2014; Sarıkaya and Güllü, 2014). Further, the analysis of variance (ANOVA) model is constructed to estimate the amount of impact of the input parameters on the outputs (Cetin et al., 2011; Camposeco-Negrete, 2013; Emami et al., 2014). However, these statistical methods hold the assumptions such as structure of a model before the problem in hand, normality of residuals, non-correlated residual error values, etc. the significant input process parameters identified using the ANOVA model is not the same for all performance characteristics (Fratila and Caizar, 2011; Hanafi et al., 2012). The reason can be attributed to the use of different material in every machining operation. Since each performance characteristic is vital, there is a need for a model that can comprehensively predict all the performance characteristics based on the given input process parameters. Thereby optimization of this model can optimize the given machining process efficiently. The models developed using such methods may not be generalized for a given input sample outside the training range. Optimization methods (Taguchi method and desirability analysis) used was the traditional ones. It can be interpreted that the research on CI methods in modelling the environmental and manufacturing characteristics has not yet attained its modernisation. Therefore, more and thorough investigation is needed to observe the influence of input parameters on the environmental aspects of the machining operations by formulation of a generalized explicit relationship between the process parameters. Alternatively, the evolutionary based multi-gene genetic programming (MGGP) which generates the model structure and its coefficients automatically can also be applied (Garg et al., 2015a,b,c). Based on the previous applications of MGGP conducted by authors (Garg et al., 2013b; Garg and Tai, 2013a,b), the complexity of the MGGP model is not accurately defined. Complexity of the evolved models during the evolutionary stages of MGGP is generally defined by number of nodes of the tree. This implies that Sin (x) and x will have same complexity as they both have 2 nodes, but it is not at all true. This is a critical issue because the complexity term is a component of the fitness function which monitors the evolutionary search and the convergence rate towards achieving the optimum solution. Therefore, determining its correct value is essentially important for the effective functioning of the algorithm by driving the evolution to its direction of global minimum. This issue also requires a thorough investigation and therefore forms a motivation for authors in developing a framework that can result in evolution of generalized models in effectively studying the impact of input process parameters on the energy consumption. In the present work, a new complexity measure based evolutionary framework of MGGP (COM-MGGP) is proposed to formulate the functional relationship between energy consumption and the three input process parameters (cutting speed, feed rate and depth of cut) of the milling process. Procedure of formulation of the energy consumption model is shown in Fig. 1. Experimental data obtained from the milling operation is further fed into the proposed evolutionary framework. In this new framework, the complexity of the models generated during the evolutionary stage is computed by number of basis functions that best fits the model. Performance of modified evolutionary framework is compared to that of standardized MGGP. The sensitivity and parametric analysis is conducted on the proposed model to understand the physics behind

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

3

Fig. 1. Procedure of formulation of energy consumption model as a function of three input process parameters.

the milling process so as to get an idea on what input parameter settings can be set for minimizing energy consumption. 2. Milling operation set-up details The details for the experimental-set up for performing the CNC face milling operation are also discussed in Yang et al. (2013). The phase of DYNA DM4600 milling machine was connected to a dual passive direct box. Milling operations were performed under dry conditions (no cutting fluid). Tool is the face milling cutter YBC301 and work piece is Cast ZG35 with dimensions 230  230  108. Experimental design approach (34) with four levels of each of the three input parameters is used to collect the data set. Cutting speed (V) is evaluated based on spindle speed by the formulae:

V¼

ND 318

(1)

where V is the cutting speed, D is the cutter diameter and n is the spindle speed. The three input process parameters considered are as follows: 1. Cutting speed (x1, metres per minute) 2. Depth of cut (x2, millimetres) 3. Feed rate (x3, millimetres per revolution) Output parameter considered is energy consumption measured in watts by energy quality analyzer. Experiments are performed in random to eliminate any effect of uncontrolled factors such as temperature, etc. CNC milling machine is programmed based on the input process parameters values. Total of 64 data samples are obtained from the milling operations. Descriptive statistics is

performed on the data to study the nature of data as shown in Table 1. These data samples are then divided into two groups a) training and b) testing group. Training group comprise of 41 data samples with the testing group of the remaining 23 data samples. Training data is used for formulating the models whereas the testing data is used for testing the generalization ability of the models. 3. New complexity measure based framework of MGGP (COMMGGP) The difference between the complexity measure based framework of MGGP (COM-MGGP) (Fig. 2) and the standardized one is the addition of a new complexity measure in its fitness function and selection of the best model based on classification strategies. The mechanism of proposed framework is listed in three steps as follows: Step 1: initial population by combining elements from functional and terminal set In the first step, the functional and terminal set is defined. The elements (arithmetic operators (þ, , /, ), non-linear functions (sin, cos, tan, exp, tanh, log) or Boolean operators) form the functional set. The elements (input variables such as suction and stress, the range of random constants) form the terminal set. The range of random constants chosen is 10 to 10. The elements from these two sets are combined randomly to form a gene. In this way, several genes are evolved. The genes are randomly chosen and regressed using the stepwise regression method to form models (individuals).

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

4

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

Table 1 Descriptive statistics of the input and output process parameters for the milling operation. Statistical Parameter

Cutting speed (x1; m/min)

Depth of cut (x2; mm)

Feed rate (x3; mm/rev)

Energy consumption (watts)

Mean Standard error Median Standard deviation Variance Kurtosis Skewness Minimum Maximum

204.19 4.43 204.1 35.50 1260.89 1.37 0.006 157 251.57

1.05 0.04 1.05 0.33 0.114 1.37 2.63 0.6 1.5

0.125 0.007 0.125 0.05 0.003 1.37 2.38 0.05 0.2

555.67 19.11 529 152.94 23393.5 0.59 0.30 274 893

Fig. 2. Flowchart showing step-by-step procedure of COM-MGGP framework.

Step 2: proposed multi-adaptive regression spline complexity term in fitness function The performance of the individuals in the initial population is evaluated based on the fitness function (structural minimization principle (SRM)) so as to avoid the over-fitting. Multi-adaptive regression splines (MARS) (Jekabsons, 2010) known for their fixed complexity (number of basis functions) are proposed for defining the complexity of the models evolved during evolutionary stages of MGGP. The mathematical representation of the MARS model is described by:

y ¼ c0 þ

N X i¼1

ci

Ki Y

  bji xvðj;iÞ

(2)

j¼1

where y is the output variable, c0 is constant, ci is vector of coefficients of the non-constant basis functions, bji(xv(j,t)) is the truncated power basis function with v(j,i) being the index of the independent variable used in the ith term of the jth product, and Ki is a parameter that limits the order of interactions. Minimum number of basis functions that best fits the model is considered as the complexity of that model. The term (b) is then

incorporated in the SRM fitness function to evaluate the performance of models during the evolutionary stages. Modified SRM function is given by:

ffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   u 0 11111 u00    u log Nb u SSE B b b b AAAC log þ@ SRM ¼ A @1  t@@  N N N N 2N 0

(3) where b is number of basis functions that best fit the model during evolutionary stages of MGGP, SSE is the sum of square of error of the generated model on the training data and N is the number of training samples. If any individual of the population does not satisfy the termination criterion, genetic operations (selection, subtree crossover and subtree mutation) are implemented on the individuals to evolve the new population. The termination criterion is the maximum number of generations or the threshold error of the model as specified by the user.

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

Step 3: selection of the best model based on classification scheme Selection of the best model is done by hybridising the classification scheme in the paradigm of proposed framework. Two classifiers, namely ANN and the Bayesian classifier, are used, which make use of metrics (training error and number of basis functions) of the generated models to form the classifiers. Selecting the appropriate structure of these classifiers is essential for its effective functioning. For ANN and Bayesian classifiers, the architecture is selected based on trial-and-error approach. The two classifiers are used so as to eliminate any sort of uncertainty in the classification accuracy of a given classifier. Notation “0” and “1” are used to classify models as “best” and “bad” respectively. After implementing several runs, the model classified as “best” by the two classifiers is selected for analysis. If there is a tie among the runs, the model of a given run with the lowest number of nodes (lower complexity) is chosen. 4. Settings for the implementation of proposed evolutionary framework The proposed evolutionary framework is implemented on the software GPTIPS (Hinchliffe et al., 1996; Searson et al., 2010) code with modification in code done for the development of graphical user interface (Table 2) for the user friendly purpose. The parameter settings of both methods (MGGP and COM-MGGP) are kept same and adjusted using a trial-and-error approach (Table 2) based on the study conducted on applications of evolutionary algorithms in manufacturing (Garg et al., 2014; Garg and Tai, 2014a,b) and in 3-D printing technology (Vijayaraghavan et al., 2015). The two methods (MGGP and COM-MGGP) are applied on the data set (data is discussed in Section 2) in the prediction of energy consumption of the milling process. The two models (MGGP and COM-MGGP, Equations (4) and (5) in Appendix) are selected. Performance and statistical comparison of these models will be discussed in Section 5. 5. Statistical evaluation of proposed models against the actual data The performance of the two methods (MGGP and COM-MGGP) is evaluated on the training and testing data (Fig. 3) in prediction of energy consumption of the milling processes. Predictions obtained are compared to those of the actual data obtained from Yang et al. (2013). Five statistical metrics (the coefficient of determination (R2), the mean absolute percentage error (MAPE), RMSE, the relative error (RE) (%) and the multiobjective error (MO)) are chosen to determine the method that gives better generalization ability. The mathematical representation of these metrics is shown below:

0

5

Pn 





12

B C Mi  Mi B C i¼1 Ai  Ai C R2 ¼ B Brffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi C     2 2 P P @ A n n A M  A  M i i i i i¼1 i¼1

(6)

  1 XAi  Mi   100  n i Ai 

(7)

MAPEð%Þ ¼

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi PN 2 i¼1 jMi  Ai j RMSE ¼ N

REð%Þ ¼

jMi  Ai j  100 Ai

Multiobjective error ¼

MAPE þ RMSE R2

(8)

(9)

(10)

where Mi and Ai are the predicted and actual values respectively, Mi and Ai are the average values of the predicted and actual respectively and n is the number of training samples. On the training phase (Fig. 3a and c) both the methods are able to learn from the data samples very well so as to form the models with high correlation coefficients and lower values of errors. On the testing phase (Fig. 3b and d), the models formed from the new proposed COM-MGGP framework has perform better than those of the other method in prediction of energy consumption. Lower values of RMSE, MAPE and the high coefficient of determination of the proposed model suggest that the predictions obtained are well in agreement with the experimental data obtained from Yang et al. (2013). MO values of the two models were computed on the training and testing data (Table 3). Descriptive statistics of the relative error (%) of the two models are shown in Table 4. It shows the several statistical metrics (mean, standard error of mean (SE error), standard deviation (Std dev), upper confidence interval (UCI), lower confidence interval (LCI), median, maximum and minimum errors) of the two models evaluated on overall data samples. The lower value of the range (range ¼ UCILCI) of confidence intervals and the remaining statistical metrics of the proposed COM-MGGP model indicates that its overall performance is better than those of the other model formed from the MGGP approach. Furthermore, the goodness of fit of the two models is computed by hypothesis testing. These are the t-tests to determine the mean and f-tests for variance. The tests are conducted on the predictions

Table 2 Settings for parameters for MGGP and COM-MGGP methods. Parameters for GP methods

Assigned values

Simulation iterations Population size/initial population of models Number of generations Tournament size Maximum depth of a gene Maximum number of genes to be combined Elements of Functional set (F) Elements of Terminal set (T) Probability of cross-over Probability of reproduction rate Probability of mutation

30 1000 300 3 6 8 (multiply, plus, minus, plog, tan, tanh, sin, cos exp) (x1, x2, x3, [10 10]) 0.85 0.10 0.05

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

6

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

Fig. 3. Graphs showing error metrics of energy consumption models (a) and (b) COM-MGGP, (c) and (d) MGGP of the milling process.

Table 3 Multi-objective error of the two energy consumption models.

Table 5 Goodness of fit of the two energy consumption models by hypothesis tests.

Models

Training phase

Testing phase

95%CI

COM-MGGP

MGGP

COM-MGGP MGGP

15.36 28.71

52.16 120.66

Mean paired t test Variance F test

0.34 0.53

0.16 0.21

obtained from the two models. For the t-tests and f-tests, the p-values of both the models >0.05 (Table 5), so there is not enough evidence to conclude that the actual values and predicted values from these models differ.

Based on the statistical evaluation of the models, it can be concluded that the model formed from the proposed COM-MGGP framework has outperformed the other and is able to capture the dynamics of the milling process in the prediction of energy consumption.

Table 4 Statistical metrics of relative error (%) of the two energy consumption models. Models

Count

Mean

LCI 95%

UCI 95%

Std dev

SE mean

Median

Maximum

Minimum

COM-MGGP MGGP

64 64

3.29 6.18

2.37 4.31

4.20 8.04

3.67 7.46

0.45 0.93

1.95 3.38

17.73 36.91

0.25 0.026

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

6. Validation of robustness of proposed models

Table 7 Impact (%) of input parameters on energy consumption of COM-MGGP model.

6.1. Validation of models on data sets generated using k-fold crossvalidation procedure This section attempts to validate the hypothesis that performance of the COM-MGGP model is better than that of the standardized MGGP model by evaluating the performance of the models on the other 4 data sets, each with different training and testing, generated using the k-fold cross-validation (k-value is chosen as 4). This is because the selection of the training and testing data set affects the learning ability of the model and thereby its generalization ability. Table 6 shows the results of the two models on the four data sets. From this, it can be concluded that COMMGGP has performed better than the standardized MGGP approach.

Table 6 Error metrics of the two models on the four data sets generated by cross-validation procedure.

Data set 1 R2 MAPE RMSE MO Data set 2 R2 MAPE RMSE MO Data set 3 R2 MAPE RMSE MO Data set 4 R2 MAPE RMSE MO

7

Models

Training phase

Testing phase

COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP

0.97 0.96 2.4 3.78 19.4 20.5 22.4 25.29

0.86 0.81 5.86 13.40 67.4 81.2 85.18 116.7

COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP

0.95 0.95 3.04 4.56 18.7 22.4 22.8 28.37

0.82 0.80 6.32 15.6 76.34 77.41 100.80 116.2

COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP

0.96 0.94 3.21 5.27 16.52 24.5 20.55 31.67

0.85 0.81 7.16 14.25 79.3 79.4 101.71 115.6

COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP COM-MGGP MGGP

0.97 0.97 2.19 3.69 14.23 15.10 16.92 19.37

0.88 0.85 5.61 11.02 53.1 59.8 66.71 83.31

Input process parameters

Impact on energy consumption

Cutting speed Feed rate Depth of cut

51.10 26.31 22.59

sensitivity (SA) and parametric analysis is to perform a check on whether the physics obtained from the COM-MGGP model matches with those from the actual understanding about the process (Yang et al., 2013). Description of the performed sensitivity and parametric analysis is given as follows: The SA percentage of the outputs to each input variable is determined using the following formulas:

Li ¼ fmax ðxi Þ  fmin ðxi Þ

(11)

L SAi ¼ Pn i

(12)

j¼1 Lj

 100

where fmax(xi) and fmin(xi) are, respectively, the maximum and minimum of the predicted output over the ith input domain, where other variables are equal to their mean values. Table 7 shows the results of sensitivity analysis of the outputs in respect to the three input variables. It is clear that the process variable, i.e. cutting speed, has the highest impact on the energy consumption followed by feed rate and depth of cut. This means that by regulating the cutting speed, the maximum variation in energy consumption can be achieved. The parametric analysis demonstrates how the outputs vary in response to the variation in input variables from their mean values. On the best COM-MGGP model, the first input is varied between its mean ± definite number of standard deviations (0.2), and the output values are computed, while the other input variables are kept fixed at their mean values. This analysis is then repeated for other inputs. Fig. 4 shows the plots generated for each input variable and the energy consumption over the range of the input variables. From these plots, the variation of energy consumption with respect to each input variable can be studied. From Fig. 4, the energy consumption increases in a parabolic style with respect to cutting speed, feed rate and depth of cut. Thus from the sensitivity and parametric analysis, we can select the optimal values of input variables which can minimize the energy consumption. In this way, the proposed model can be used to conserve greater amount of energy in the milling operation. 6.3. Comparison of performance of COM-MGGP model with that of ANOVA and analytical model

6.2. Study of relationships of energy consumption and input parameters via sensitivity and parametric analysis of the COMMGGP model From Section 6.1, it can be concluded that the COM-MGGP model is the best. The main reason behind performing the

6.3.1. Comparison with ANOVA model The performance of the COM-MGGP model is compared to that of the ANOVA model. ANOVA model on 4-levels of each input is formulated. Model (Equation (13)) considered was of second order with interaction terms.

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

8

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

ANOVA Model for Energy Consumption:

Though the performance of the ANOVA and COM-MGGP models is similar, the ANOVA model holds several statistical assumptions (Fig. 5) such as the pre-defined structure of the model, independent and un-correlated residuals and is valid only in the given input range, i.e. cannot be extrapolated. On the other side, COM-MGGP model possesses the self ability of forming nonlinear curve with any number/order of interaction terms on the given data, which can be further extrapolated. This is supported by the fact the COM-MGGP model being formulated on 41 data samples is able to generate correlation coefficient of almost 0.90, which is the same as what the ANOVA model formulated on 64 samples. Comparing the results of sensitivity and parametric analysis, the similar inference of cutting speed as the most dominant parameter followed by feed rate and depth of cut is found. 6.3.2. Comparison with analytical model The performance of the empirical model (Armarego et al., 2000) being developed by estimating its coefficients is compared to that of the COM-MGGP model. The mathematical representation of the analytical model is given by

P ¼ Cfta abp aεr Dg Vz

(14)

where C, a, g, b and ε are constants, D is the milling diameter, ar is the depth of cut, z is the tooth number of milling cutter, V is the cutting speed, ft is the feed rate and ap is the depth of cut. With ar, D and z as constants the expression is simplified to

P ¼ Cfta abp V

(15)

The expression is fitted using the least squares method on the data given in Table 1. The formulate estimated is

P ¼ ft0:22 *a0:27 *4:4*V p

(16)

MAPE of the model (Equation (16)) is found to be 8.45 on the testing data, which is higher than any of the MAPE of COMMGGP model given in Table 5. Based on this interpretation, it can be concluded that the model formed from the proposed COM-MGGP model has performed better than that of the analytical model.

7. Conclusions From the perspective of energy conservation of milling process, the present work emphasizes the need for the formulation of generalized functional expressions. In addition, the generalization issue in functioning of MGGP is overcome by the formulation of a new complexity measure based evolutionary framework (COM-MGGP). This framework incorporates the use of MARS for computing the complexity of every model evolved during evolutionary stages of algorithm. The higher generalization ability of the proposed model obtained is beneficial for reducing the energy consumption of the milling process by

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

Fig. 4. Plots showing the variation of energy consumption with respect to the three input process parameters.

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

9

10

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

Fig. 5. Four in one plots for validation of statistical assumptions of the ANOVA model.

predicting it in uncertain input process conditions and thus avoiding the use of a trial-and-error approach. The conducted sensitivity and parametric analysis validates the robustness of the COM-MGGP model by unveiling the hidden relationships of energy consumption with respect to a set of input variables. From these relationships, the appropriate input parameter settings that minimize the energy consumption can be obtained. This also implies that the greater amount of energy can be conserved. The difference between the minimum energy consumption achieved when optimized in case of proposed and standardized framework is the energy conserved because the proposed framework when optimized would result in determination of accurate optimum values of speed, feed rate and depth of cut in achieving a global minimum value of energy consumption. Other advantages of using the proposed approach over the traditional methods are that it possesses an inherent ability of selecting the relevant input variables that contribute significantly to the outputs of the milling process. Thus, the use of ANOVA can be avoided.

Energy consumption

MGGP

The current work is limited to one objective (energy consumption) and therefore the future work for the authors is to focus on multi-objective optimization of both the manufacturing (surface roughness and tool life) and environmental aspects (Lam, 2015) of the milling process simultaneously by using the robust genetic algorithms.

Acknowledgments The study was supported by the Singapore MPLP project, Nanyang Technological University ref. M4061473 and Research Grant ref. M060030008.

Appendix

¼ 296:2974 þ ð173:564Þ*ðx2Þ þ ð2737:0799Þ*ðððtanðx3ÞÞ*ðtanðcosðsinðx2ÞÞÞÞÞ*ðx2ÞÞ þ ð0:0057986Þ*ðððx1Þ*ðtanðððx3Þ þ ðx2ÞÞ þ ðx3ÞÞÞÞ*ðx2ÞÞ þ ð1277:8907Þ*ðx3Þ þ ð1570:0168Þ*ðsinððtanhðððx2Þ*ðx3ÞÞ*ðsinðx3ÞÞÞÞ*ððtanððx1Þ*ðx1ÞÞÞÞÞÞ þ ð3402:4127Þ*ðtanðððx2Þ*ðtanðx1ÞÞÞ*ðsinðx3ÞÞÞÞ þ ð2:8011Þ*ðððððx2Þ*ðx3ÞÞ*ðsinðx3ÞÞÞ

(4)

þ ðx3ÞÞ  ðððx1Þ þ ðsinðx3ÞÞÞ þ ðtanðððx3Þ þ ðx1ÞÞ þ ðx3ÞÞÞÞÞ;

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

11

Energy consumption COMMGGP ¼ 77:8022 þ ð  659:8778Þ*ðcosððtanhðððx1Þ*ðð  0:036115ÞÞÞ þ ðsinðx1ÞÞÞÞ  ððtanððx3Þ*ðx2ÞÞÞ þ ðtanðcosðx3ÞÞÞÞÞÞ þ ð  0:0023738Þ*ðððtanððsinðx3ÞÞ þ ððx2Þ þ ðx3ÞÞÞÞ  ððððx2Þ*ðx1ÞÞ*ððx3Þ*ðx2ÞÞÞ  ðtanhðsinðð8:052523ÞÞÞÞÞÞ*ððx1Þ  ðcosðcosðtanhðx1ÞÞÞÞÞÞ þ ð  0:16278Þ*ðððtanððsinðx3ÞÞ þ ððx3Þ*ðx2ÞÞÞÞ  ððððx2Þ*ðx1ÞÞ*ððx3Þ*ðx1ÞÞÞ  ðtanhðsinðð8:052523ÞÞÞÞÞÞ*ððcosðx3ÞÞ  ðcosðcosðtanhðx1ÞÞÞÞÞÞ þ ð1323:0989Þ*ðcosððtanhðððx1Þ*ðð  0:036115ÞÞÞ þ ðtanhðx2ÞÞÞÞ  ððtanððx3Þ*ðx3ÞÞÞ þ ðtanhððx2Þ*ðx1ÞÞÞÞÞÞ þ ð  158:3252Þ*ðððcosðððx1Þ*ðð  0:036115ÞÞÞ þ ððx3Þ þ ðx1ÞÞÞÞ  ððððx2Þ  ðx3ÞÞ*ððx3Þ*ðx2ÞÞÞ  ðtanhðsinðð8:052523ÞÞÞÞÞÞ*ððcosððsinðx2ÞÞ þ ððx3Þ þ ðx3ÞÞÞÞ  ðcosðcosðtanhðx1ÞÞÞÞÞÞ þ ð0:0083412Þ*ððððx1Þ þ ððð  4:780656ÞÞ*ðð9:203756ÞÞÞÞ*ðx1ÞÞ  ðtanhðsinðx2ÞÞÞÞ þ ð  1:0786Þ*ðððtanðððx1Þ þ ðx2ÞÞ þ ððð0:287999ÞÞ þ ðx3ÞÞÞÞ  ððððx2Þ  ðx3ÞÞ*ððx3Þ*ðx2ÞÞÞ  ðtanhðsinðx3ÞÞÞÞÞ*ððcosðððx1Þ þ ðx3ÞÞ*ðcosðx2ÞÞÞÞ  ðcosðcosðtanhðx1ÞÞÞÞÞÞ (5)

References Ahilan, C., Kumanan, S., Sivakumaran, N., Edwin Raja Dhas, J., 2012. Modeling and prediction of machining quality in CNC turning process using intelligent hybrid decision making tools. Appl. Soft Comput. 13 (3), 1543e1551. http://dx.doi.org/ 10.1016/j.asoc.2012.03.071. Armarego, E.J.A., Ostafiev, D., Wong, S.W.Y., Verezub, S., 2000. An appraisal of empirical modeling and proprietary software databases for performance prediction of machining operations. Mach. Sci. Technol. 4 (3), 479e510. http:// dx.doi.org/10.1080/10940340008945719. Aykut, C.S., Golc, U.M., Semiz, S., Ergur, H.S., 2007. Modeling of cutting forces as function of cutting parameters for face milling of satellite 6 using an artificial neural network. J. Mater. Process Tech. 190 (1), 199e203. http://dx.doi.org/ 10.1016/j.jmatprotec.2007.02.045. Balogun, V.A., Mativenga, P.T., 2013. Modelling of direct energy requirements in mechanical machining processes. J. Clean. Prod. 41, 179e186. Bhushan, R.K., 2013. Optimization of cutting parameters for minimizing power consumption and maximizing tool life during machining of Al Alloy Sic particle composites. J. Clean. Prod. 39, 242e254. Campatelli, G., Lorenzini, L., Scippa, A., 2014. Optimization of process parameters using a response surface method for minimizing power consumption in the milling of carbon steel. J. Clean. Prod. 66, 309e316. Camposeco-Negrete, C., 2013. Optimization of cutting parameters for minimizing energy consumption in turning of Aisi 6061 T6 using Taguchi methodology and Anova. J. Clean. Prod. 53, 195e203. Cetin, M.H., Ozcelik, B., Kuram, E., Demirbas, E., 2011. Evaluation of vegetable based cutting fluids with extreme pressure and cutting parameters in turning of Aisi 304l by Taguchi method. J. Clean. Prod. 19, 2049e2056. Chandrasekaran, M., Muralidhar, M., Krishna, C.M., Dixit, U., 2010. Application of soft computing techniques in machining performance prediction and optimization: a literature review. Int. J. Adv. Manuf. Technol. 46, 445e464. Davim, J.P. (Ed.), 2011. Machining of Hard Materials. Springer Science & Business Media. Egilmez, G., Kucukvar, M., Tatari, O., Bhutta, M.K.S., 2014a. Erratum: supply chain sustainability assessment of the U.S. food manufacturing sectors: a life cyclebased frontier approach (Resources, Conservation and Recycling (2014) 82 (820)). Resour. Conserv. Recycl. 86, 147e148. Egilmez, G., Kucukvar, M., Tatari, O., Bhutta, M.K.S., 2014b. Supply chain sustainability assessment of the U.S. food manufacturing sectors: a life cycle-based frontier approach. Resour. Conserv. Recycl. 82, 8e20. Emami, M., Sadeghi, M.H., Sarhan, A.A.D., Hasani, F., 2014. Investigating the minimum quantity lubrication in grinding of Al2 O3 engineering ceramic. J. Clean. Prod. 66, 632e643. Hinchliffe, M., Hiden, H., Mckay, B., Willis, M., Tham, M., Barton, G., Koza, J.R., 1996. Modelling chemical process systems using a multi-gene. Late Breaking Papers at the Genetic Programming, pp. 56e65. Li, X.P., Zheng, H.Q., Wong, Y.S., Nee, A., 2000. An approach to theoretical modeling and simulation of face milling forces. J. Manuf. Process. 2 (4), 225e240. http:// dx.doi.org/10.1016/S1526-6125(00)70024-7. Li, W., Zein, A., Kara, S., Herrmann, C., 2011. Glocalized solutions for sustainability in manufacturing. In: Hesselbach, J., Herrmann, C. (Eds.), Glocalized Solutions for Sustainability in Manufacturing: Proceedings of the 18th CIRP International Conference on Life Cycle Engineering. Springer-Verlag, Berlin Heidelberg, Germany, pp. 268e273. Fang, K., Uhan, N., Zhao, F., Sutherland, J.W., 2011. A new approach to scheduling in manufacturing for power consumption and carbon footprint reduction. J. Manuf. Syst. 30, 234e240. Fratila, D., Caizar, C., 2011. Application of Taguchi method to selection of optimal lubrication and cutting conditions in face milling of Almg3. J. Clean. Prod. 19, 640e645.

Garg, A., Lam, J.S.L., 2015. Improving environmental sustainability by formulation of generalized power consumption models using an ensemble based multi-gene genetic programming approach. J. Clean. Prod. 102, 246e263. Garg, A., Tai, K., 2013a. Selection of a robust experimental design for the effective modeling of nonlinear systems using genetic programming. In: Computational Intelligence and Data Mining (Cidm), 2013c IEEE Symposium. IEEE, pp. 287e292. Garg, A., Tai, K., 2013b. Genetic programming for modeling vibratory finishing process: role of experimental designs and fitness functions. In: Swarm, Evolutionary, and Memetic Computing. Springer International Publishing, pp. 23e31. Garg, A., Tai, K., 2014a. An ensemble approach of machine learning in evaluation of mechanical property of the rapid prototyping fabricated prototype. Appl. Mech. Mater. 575, 493e496. Garg, A., Tai, K., 2014b. Stepwise approach for the evolution of generalized genetic programming model in prediction of surface finish of the turning process. Adv. Eng. Softw. 78, 16e27. Garg, A., Bhalerao, Y., Tai, K., 2013a. Review of empirical modelling techniques for modelling of turning process. Int. J. Model. Identif. Control 20, 121e129. Garg, A., Sriram, S., Tai, K., 2013b. Empirical analysis of model selection criteria for genetic programming in modeling of time series system. In: Computational Intelligence for Financial Engineering & Economics (Cifer), 2013b IEEE Conference. IEEE, pp. 90e94. Garg, A., et al., 2014. Combined CI-MD approach in formulation of engineering moduli of single layer graphene sheet. Simul. Model. Pract. Theory 48, 93e111. Garg, A., Vijayaraghavan, V., Lam, J.S.L., Singru, P.M., Gao, L., 2015a. A molecular simulation based computational intelligence study of a nano-machining process with implications on its environmental performance. Swarm Evol. Comput. 21, 54e63. Garg, A., Lam, J.S.L., Savalani, M.M., 2015b. A new computational intelligence approach in formulation of functional relationship of open porosity of the additive manufacturing process. Int. J. Adv. Manuf. Technol. http://dx.doi.org/ 10.1007/s00170-015-6989-2. Garg, A., Garg, A., Lam, J.S.L., 2015c. Evolving functional expression of permeability of fly ash by a new evolutionary approach. Transp. Porous Media 107 (2), 555e571. Hanafi, I., Khamlichi, A., Cabrera, F.M., Almansa, E., Jabbouri, A., 2012. Optimization of cutting conditions for sustainable machining of Peek-Cf30 using tin tools. J. Clean. Prod. 33, 1e9. He, Y., Liu, B., Zhang, X., Gao, H., Liu, X., 2012. A modeling method of task-oriented energy consumption for machining manufacturing system. J. Clean. Prod. 23, 167e174. Jekabsons, G., 2010. M5Primelab: M5'regression Tree and Model Tree Toolbox for Matlab. Octave. Institute of Applied Computer Systems. Kant, G., Sangwan, K.S., 2014. Prediction and optimization of machining parameters for minimizing power consumption and surface roughness in machining. J. Clean. Prod. 83, 151e164. Kreiger, M.A., Shonnard, D.R., Pearce, J.M., 2013. Life cycle analysis of silane recycling in amorphous silicon-based solar photovoltaic manufacturing. Resour. Conserv. Recycl. 70, 44e49. Lam, J.S.L., 2015. Designing a sustainable maritime supply chain: a hybrid QFD-ANP approach. Transp. Res. Part E 78, 70e81. Lam, J.S.L., Lai, K.H., 2014. Developing environmental sustainability by ANP-QFD approach: the case of shipping operations. J. Clean. Prod. http://dx.doi.org/ 10.1016/j.jclepro.2014.09.070. pez de Lacalle, L.N., Lamikiz, A., Sanchez, J.A., de Bustos, I.F., 2005. Simultaneous Lo measurement of forces and machine tool position for diagnostic of machining tests. Instrum. Meas. IEEE Trans. 54 (6), 2329e2335. pez de Lacalle, L.N., Lamikiz, A., Sa nchez, J.A., de Bustos, I.F., 2006. Recording of Lo real cutting forces along the milling of complex parts. Mechatronics 16 (1), 21e32.

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043

12

A. Garg et al. / Journal of Cleaner Production xxx (2015) 1e12

Sarıkaya, M., Güllü, A., 2014. Taguchi design and response surface methodology based analysis of machining parameters in CNC turning under MQL. J. Clean. Prod. 65, 604e616. Searson, D.P., Leahy, D.E., Willis, M.J., 2010. Gptips: an open source genetic programming toolbox for multigene symbolic regression. In: Proceedings of the International Multiconference of Engineers and Computer Scientists. Citeseer, pp. 77e80. Siddique, R., 2012. Utilization of wood ash in concrete manufacturing. Resour. Conserv. Recycl. 67, 27e33. De Soete, W., Boone, L., Willemse, F., De Meyer, E., Heirman, B., Van Langenhove, H., Dewulf, J., 2014. Environmental resource footprinting of drug manufacturing: effects of scale-up and tablet dosage. Resour. Conserv. Recycl. 91, 82e88. Tat ano, F., Acerbi, N., Monterubbiano, C., Pretelli, S., Tombari, L., Mangani, F., 2012. Shoe manufacturing wastes: characterisation of properties and recovery options. Resour. Conserv. Recycl. 66, 66e75.

Tolio, T., Sacco, M., Terkaj, W., Urgo, M., 2013. Virtual factory: an integrated framework for manufacturing systems design and analysis. Proced. CIRP 7, 25e30. Vijayaraghavan, V., Garg, A., Lam, J.S.L., Panda, B., Mahapatra, S.S., 2015. Process characterization of 3D-printed FDM components using improved evolutionary computational approach. Int. J. Adv. Manuf. Technol. 78 (5), 781e793. Yan, J., Li, L., 2013. Multi-objective optimization of milling parameterseThe tradeoffs between energy, production rate and cutting quality. J. Clean. Prod. 52, 462e471. Yang, Y., Li, X., Gao, L., Shao, X., 2013. Modeling and impact factors analyzing of energy consumption in CNC face milling using GRASP gene expression programming. Int. J. Adv. Manuf. Technol. 1e17. http://dx.doi.org/10.1007/s00170013-5017-7.  pez de Lacalle, L.N., 2011. An integrated processemachine Zulaika, J.J., Campa, F.J., Lo approach for designing productive and lightweight milling machines. Int. J. Mach. Tools Manuf. 51 (7), 591e604.

Please cite this article in press as: Garg, A., et al., Energy conservation in manufacturing operations: modelling the milling process by a new complexity-based evolutionary approach, Journal of Cleaner Production (2015), http://dx.doi.org/10.1016/j.jclepro.2015.06.043