Measurement 47 (2014) 78–91

Contents lists available at ScienceDirect

Measurement

journal homepage: www.elsevier .com/ locate /measurement

Enhance performance of inspection process on CoordinateMeasuring Machine

0263-2241/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.measurement.2013.08.045

⇑ Corresponding author. Mobile: +966 553118695.E-mail address: syedhammad68@yahoo.co.in (S.H. Mian).

Syed Hammad Mian ⇑, Abdulrahman Al-AhmariAdvanced Manufacturing Institute, College of Engineering, King Saud University, P.Box 800, Riyadh 11421, Saudi Arabia

a r t i c l e i n f o a b s t r a c t

Article history:Received 23 February 2013Received in revised form 4 August 2013Accepted 19 August 2013Available online 28 August 2013

Keywords:Coordinate Measuring Machine (CMM)Measurement sequenceInspectionOptimization plotBrute force (BF)Genetic algorithm (GA)Simulated algorithm (SA)

Coordinate Measuring Machine (CMM) has been an important inspection tool in qualitycontrol for several years owing to its high accuracy and precision. Effectiveness of inspec-tion plan generated by CMM greatly depends on measurement cycle time. Lesser theinspection time taken by CMM to measure a given part better will be the performance ofinspection process. Therefore, it has been critical to reduce measurement time for efficientperformance of inspection process. In this paper, methodologies to generate most suitablemeasurement path resulting into minimum inspection time has been introduced. Thesemethodologies are based on different algorithms to reduce measurement cycle time forCMM. The different algorithms have successfully been explored and compared to showtheir effectiveness in minimizing inspection time for stationary CMM equipped with touchtrigger probe. The proposed methodologies have also been implemented and tested onreal-world mechanical part with certain number of features to demonstrate theirapplicability.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Coordinate Measuring Machine (CMM) equipped withcontact probe has been standard and most frequently usedmeasuring instrument for dimensional inspection. Regard-less of the availability of large number of non-contact mea-suring devices, CMMs mounted with touch probe havebeen preferred choice for inspection purposes. This is dueto the fact that it can offer very high accuracy dependingon the environment within which it operates. Since, CMMsrequires huge capital investment therefore their properutilization has been primary concern in industries. More-over, ever increasing demand of high quality componentsand stiff competition in market requires manufacturers toreduce inspection time without compromising inspectionquality. It becomes even more important to speed upinspection process on manufacturing line when number

of features being measured increases. CMM probe travelsto various features on the part depending on inspectionplan during the inspection process. Therefore,determination of most appropriate inspection path hasbeen critical for improved performance of CMMs. In anyinspection plan, CMM measurement probe can travel tomeasure any feature but efficient inspection plan resultsin measurement sequence that minimizes inspection time.Thus, effective inspection path provides shortest route tomeasure number of features on the part being inspected.Overall improvement of CMM inspection process can alsoinclude minimizing number of probe re-orientation, reduc-tion of stand-off distance (clearance, retract, approachdistances etc.) etc., during inspection run besidesmeasurement sequence. However, in present work, onlymeasurement sequence has been considered because it ismost critical aspect and contributes significantly to theperformance of CMM. The motivation for this work hasbeen provided by demands of effective methods and tech-niques that can be used to improve efficiency of inspectionprocess thus reducing inspection time.

S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91 79

The objective of this research includes determination ofbest possible probing sequence to measure different fea-tures on part. This problem requires generation of idealpath that measurement probe should take while movingbetween different features. The work at hand focuses onapplication of various algorithms and their comparisonsto minimize measurement time. Although, applications ofCMM for inspection purposes has extensively been spreadthroughout manufacturing industries yet measurement se-quence planning especially by using heuristics such as ge-netic algorithm (GA), simulated annealing (SA) and bruteforce (BF) has not fully been developed. Therefore, in thiswork, methodologies to implement well established tech-niques such as GA, SA, and BF which have already shownmany successful manufacturing applications have beenproposed for improvement of inspection process.

2. Literature review

A large amount of work is being carried out to improveperformance of CMM inspection process owing to in-creased demands of shorter inspection time. According toTopfer et al. [1] CMM performance can be improved withoptimum measurement strategy involving minimal tra-verse path, minimal measuring time and minimal degreeof wear. However, main objective has always beenimprovement of measurement path to reduce inspectiontime. There have been many techniques that can be ex-plored and implemented to reduce measurement time forcomplex parts made up of several features. For instance,Ant Colony Optimization (ACO) based algorithm by Ji andLi [2] optimized measurement path for digitization ofsupercharger impeller. Similarly, principle of GA has suc-cessfully been implemented by Lin et al. [3] to generatemeasurement path over prismatic polyhedral parts. Theproblems of parameter selection, different operator’sbehavior, premature convergence problem etc., in GA havecomprehensively been studied by Qu et al. [4] to optimizeCMM measurement path. Approximation algorithms suchas nearest neighbor method and refinement method byLin and Chen [5] have also identified measurement se-quence for better performance of CMM inspection process.Lu et al. [6] successfully achieved optimization of CMMinspection path and reduced inspection time considerably.Moreover, Buchal and Wang [7] applied nearest-neighborTravelling Salesman Problem (TSP) algorithm to find outbest possible inspection sequence. The need of shorterinspection time and better inspection methods have beendriving force behind increased attempts to minimizeinspection time. Implementation of artificial neuralnetwork (ANN) by Ruegsegger [8] produced satisfactory re-sults with optimized measurement sequence of inspectionpoints. Likewise, application of ANN by Lu et al. [9] to opti-mize inspection path resulted in reduced measurementtime for multiple component inspection process.

Techniques such as GA, SA, BF, on account of theirnumerous benefits have been finding many applicationsin manufacturing industries. Successful application of GAby Cus and Balic [10] to determine cutting parameters foroptimized machining conditions has proved that GA based

optimization methods are robust, effective and efficient.Therefore, they can be used for variety of complex optimi-zation problems. Similarly, Kolahan and Khajavi [11] opti-mized cutting parameters for abrasive water jet machiningusing SA and suggested SA an efficient and effective solu-tion for optimization problems. GA has also been used byKaya [12] for its flexibility and effectiveness in order tooptimize fixture layout problem. Like, Huang et al. [13]who introduced GA for sequencing of operations in weld-ing process, Qudeiri et al. [14] also employed GA to findbest sequence of operations with minimum tool travelpath for CNC machining operation. Moreover, Low et al.[15] minimized traveling path of gantry robot throughoptimization using GA whereas Garg and Kumar [16] pre-sented and compared optimization capabilities of GA andSA for path planning of robotic manipulators.

Geng et al. [17] proposed algorithm based on SA andgreedy search techniques to solve TSP and demonstratedits effectiveness with number of TSP instances. Similarly,Baraglia et al. [18] employed hybrid algorithm combiningGA and Lin-Kernighan (LK) local search heuristic for solv-ing TSP and showed its effectiveness through experimentson various TSP instances. Chen and Chien [19] also usedcombination of SA, GA and ACO along with particle swarmoptimization technique to solve TSP. Furthermore, Zichenget al. [20] presented two stage SA methods to optimize TSP.First stage produced initial solution using simple SA whilesecond stage utilized an effective SA to produce optimizedsolution for TSP. The effectiveness of this algorithm wasconfirmed through its application on number of bench-mark TSP instances. As far as BF approach is concerned,researchers have been working to explore its potential inmanufacturing applications. Meanwhile, BF approach hasfound several applications in machine scheduling problemand TSP. For example, Kodeekha [21] implemented BFmethod to Flexible Manufacturing System (FMS) in orderto provide effective scheduling solution. BF methodprovides simplest, quickest and efficient method tooptimize travelling path and minimize travelling time forTSP [22].

Following conclusions can be made based on literaturereview:

� Widespread applications of SA and GA have proved thatalgorithms based on SA and GA can produce qualitativeand superior results in lesser computational time.� Limited research work has been found with respect to

BF approach in manufacturing applications.� Problem of sequencing measurement features for part

inspection is similar to TSP. Therefore, sequencing offeatures for part inspection can be formulated as stan-dard TSP. Features on part can be considered as citieswhile distance between features is similar to distancetravelled between two cities of TSP. Moreover, objectivefunction in both cases is similar that is minimization ofmeasurement time for inspection planning and travel-ling time in case of TSP.� Measurement path during inspection process can be

determined by measurement sequence whereas lengthof measurement path determines total measurementdistance and hence measurement time.

80 S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91

Moreover, contributions of this work include:

� Three different algorithms have been considered toobtain measurement sequence that minimizes CMMinspection time. Algorithms based on improved heuris-tics i.e. GA and SA and algorithm based on exactapproach such as BF approach have been presented tominimize inspection time.� In this paper, adjustment of different parameters has

been carried out to improve performance of GA andSA based algorithms and to get best solution in mini-mum time.� Application of these algorithms has been verified and

compared through inspection of real world mechanicalpart.

3. Methodology

The objective of this work requires determination ofeffective measurement path that can minimize inspectiontime. Measurement time for inspection process differsdepending on the length of measurement sequence. There-fore, objective (fitness) function for this problem can be de-fined as follows:

E ¼Xn

i¼1

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðxi � xiþ1Þ2 þ ðyi � yiþ1Þ

2 þ ðzi � ziþ1Þ2q

where n defines number of measuring features, (xi,yi,zi)represent coordinates of features i.e. their location on thepart and E is the length of measurement sequence i.e. totaldistance travelled by CMM probe through the part. Mea-surement sequence that can minimize value E actually rep-resent effective probe path resulting in minimummeasurement time. Proposed algorithms take number offeatures and their corresponding coordinates as input todetermine shortest probing path. The working and imple-mentation of different algorithms have been explained infollowing sections.

3.1. Algorithm based on GA

GA can be defined as global search heuristics that workson the principle of natural evolution. It is an iterative pro-cedure where large population of individual solutions(candidate solutions) is maintained. GA is capable of find-ing global or near-global optimum solutions for multi-objective functions.

Improvement of inspection path in order to get shortestmeasurement time is a single objective problem whereonly one function (i.e. Euclidean distance) needs to be opti-mized. For each individual (measurement sequence) in thepopulation, fitness value (i.e. Euclidean distance) has to beevaluated. This fitness value of each individual is used inlater stages of algorithm when selection, crossover andmutation operators have to be applied for generation ofnew population. Methodology for application of GA basedalgorithms as shown in Fig. 1 can be described as follows.

1. Generation of initial population: Execution of this algo-rithm requires set of measurement sequences. There-fore, first step creates a set of individuals(measurement sequences) to represent initial popula-tion. Initial population in this problem has been gener-ated using random numbers. Then, each individual inthe given population has to be evaluated using their fit-ness values.

2. Encoding: Individuals (chromosomes) in GA based algo-rithms are constituted by set of genes which can be rep-resented either as integers, boolean or string variablesetc. Encoding of chromosome has been one of the vitalsteps in GA which mainly depends on the nature ofproblem to be solved. Since, objective of present prob-lem requires determination of optimum measurementsequence therefore permutation encoding has beenselected. In this type of coding, every measurementsequence (chromosome) has to be represented by stringof numbers such as 1 4 6 2 5 8 and so on where eachnumber (1 4 6 2 5 8) represents a feature. For example,1 represents feature 1, 2 represents feature 2 and so on.Subsequent steps in this algorithm require applicationof evolutionary operators such as selection, crossoverand mutation to produce new set of more fitindividuals.

3. Selection: of individuals from a given set of populationdepends on their fitness values. Since, given optimiza-tion problem is a minimization problem therefore, indi-viduals with minimum fitness would have morechances to be selected for next generation. There havebeen many methods for selection in GA but keeping inmind the nature of problem and depending on theresults of preliminary experiments, rank selection [23]has been chosen as selection technique for this prob-lem. Rank selection technique is based on selectionprobability which is proportional to relative fitnessrather than absolute fitness of individuals. Rank selec-tion overcomes the limitation of roulette wheel selec-tion method where majority of individuals areneglected. In this problem, individuals in the populationhave been ranked based on their fitness and then mea-surement sequence with minimum fitness value i.e. oneinvolving minimum distance of CMM probe has beenselected. The next important step in this algorithminvolves implementation of ‘‘crossover and mutation’’operators. The performance and effectiveness of anyGA based algorithm greatly depends on application ofthese two basic operators.

4. Crossover: Crossover operation also called as recombi-nation involves interchanging of genes between twoparent chromosomes (individuals) to produce anentirely two new set of child chromosomes. Ordinarycrossovers such as single point crossover and two pointcrossovers are not applicable for this problem. This isbecause measurement sequence (chromosome) hasbeen represented by string of numbers using permuta-tion encoding technique. As a result of permutationencoding, operation of ordinary crossovers in manycases generate child chromosomes (new measurementsequence) that are invalid with some features missingand some repeated in measurement sequence. There

Fig. 1. Implementation of algorithm based on GA to obtain measurement sequence.

S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91 81

have been many techniques throughout literature toovercome this kind of issue but keeping in mind thenature of present problem, specialized crossover inthe form of ‘‘order crossover’’ [24,25] has been selected.Application of order crossover provides several benefitsowing to its simplicity, ease and convenience to givenproblem. The objective behind application of special-ized crossover is to create valid chromosomes ratherthan generating any invalid chromosome. Functioningof ‘‘order crossover’’ for given problem can be describedas follows:

1. Identify crossover point e.g. one crossover point hasbeen identified at 3 in Fig. 2 and two crossover pointshave been selected at 3 and 6 in Fig. 3.

2. To generate first child i.e. new measurement sequence 1(nm1), elements up to crossover point in measurementsequence (m1) have been copied at correspondingpoints in nm1 as shown in Fig. 2. The remaining pointsin nm1 have been copied from the starting of parent 2which is measurement sequence 2 (m2) in their respec-

tive order but elements that already exists in nm1 havebeen skipped. Similarly, in case of two crossover points,elements between two crossover points in parent 1(m1) have been copied at corresponding points in child(nm1) whereas remaining elements have been copiedfrom the starting of m2 in order in which these ele-ments exist in m2. The elements that already exist innm1 have not been copied from m2 to avoid any repe-tition as shown in Fig. 3.

3. To generate second child i.e. nm2, elements up to cross-over point in m2 have been copied at correspondingpoints in nm2. The remaining points in nm2 have beencopied from the starting of m1 in their respective orderbut avoiding those that already exist in nm2. Similarly,in two crossover points, elements between two cross-over points in m2 have been copied at the correspond-ing points in nm2 whereas remaining elements havebeen copied from the starting of m1 in the same orderas in m1. The elements that already exist in nm2 havenot been copied from m1 in order to avoid repetition.

Fig. 2. Operation of order crossover when one crossover point is identified.

Fig. 3. Operation of order crossover when two crossover points are selected.

Fig. 4. Functioning of exchange mutation.

82 S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91

The advantage with this crossover is that new measure-ment sequence is always different from parent measure-ment sequence.

5. Mutation: Mutation involves random interchanging ofgenes for individuals in the population. Mutation oper-ator randomly interchanges positions in the sequencei.e. it randomly swaps positions of features in randomlyselected measurement sequence. For this problem,exchange mutation [26] operator has been used. Thisoperator works by randomly selecting two positionsin the sequence and then swapping the correspondingelements as shown in Fig. 4. This mutation operator is

simple, easy to implement for given problem andimproves variability of population by introducing newindividuals.

6. Termination: This algorithm terminates when numberof generated individuals in the population equals thespecified number. Therefore, after pre-specified numberof iterations i.e. when number of iterations in the prob-lem reaches maximum number of iterations, programstops. This stopping criterion has been used because itensures results within sometime whether algorithmreaches extreme or not. Moreover, this stoppingcriterion has resulted in lesser CPU time. Fresh searchcan be initiated if no acceptable solution is reached.

Fig. 5. Pseudo code for generating GA based algorithm.

S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91 83

Main problem associated with execution of GA basedalgorithm has been concerned with appropriate settingsfor GA parameters. GA parameters include crossover rate,mutation rate, population size and number of iterations(stopping condition). To overcome this issue, it has beendecided to utilize Design of experiment (DOE) techniquefor generating design conditions. Design conditions basedon full-factorial have been utilized to identify all combina-tions of various levels for GA parameters. After successfulexecution of GA at different combinations, results have tobe analyzed for best GA parameter settings. Pseudo codefor implementing GA based algorithm has been presentedin Fig. 5. Moreover, different GA parameters can be definedas follows:

� Population size: It represents number of individuals(chromosomes) in the population.� Crossover rate: Crossover rate is defined as the ratio of

number of children produced in the population to pop-ulation size [27].� Mutate rate: It represents percentage of individuals that

mutates. Mutation rate controls the rate at which newindividuals are generated in the population [27]. Theprobabilities for crossover and bit mutation typicallyrange over 0.6–0.95 and 0.001–0.01 respectively [28].

3.2. Algorithm based on SA

SA has been an important tool for finding acceptablesolution to complex problems with large number of possi-ble solutions [29]. Major advantage with SA is that it avoidstrapping in local optimum and obtains good quality solu-tions. It is capable of finding optimum or near optimumsolutions to wide variety of problems. The working princi-ple of SA is analogous to the concept of physical annealingwhere metal is first heated to very high temperature andthen slowly cools down to get desired structural proper-ties. For given problem, this algorithm begins by definingcoordinates for part features and determining correspond-ing distances between all defined features.

Methodology (shown in Fig. 6) for application ofSA based algorithm have been described as follows.

However, pseudo code for SA based algorithm can be seenin Fig. 8.

1. Input coordinates of all features on the part.2. Calculate distance matrix to define distances between

each feature.3. Generate initial solution randomly such as 3 5 2 6 1 4.

Numbers 1 2 3 4 5 6 defines different features and 3 52 6 1 4 represents measurement sequence.

4. Set appropriate values for the following SA parameters:

� Initial high temperature T.� Cooling factor.

� Maximum number of iterations.� Stopping tolerance.

Appropriate values for SA parameters have been ob-tained using DOE technique. The procedure of identifyingsuitable values for different SA parameters begins bydeciding number of levels and value for each level of allparameters. Design table defining different experimentalconditions has been generated using full factorial design.SA loop has been run for different conditions and resultsthus obtained have been analyzed to get appropriate val-ues. These appropriate values have then finally been usedto accomplish desired objective.

5. SA loop begins by randomly generating initialmeasurement sequence and then finding its totaldistance using distance matrix.

6. Subsequent stage requires generation of neighbor-hood solution i.e. new measurement sequence. Inpresent case, neighborhood measurement sequencehas been generated by first generating three posi-tions randomly and then transposing (reversing) thesequence of features at those generated positions asshown in Fig. 7.

7. Energy states i.e. total distance moved by CMMprobe has been computed for both initial solution(measurement sequence) and neighborhood solution(new measurement sequence) using distance matrix.

8. Once neighborhood solution has been generated, ithas to be checked for its acceptance based on accep-tance criterion (c < D).

Fig. 6. Algorithm based on SA to obtain best measurement sequence.

84 S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91

Where D ¼ eðd1�d2 Þ

T and c is the random number in theinterval (0,1)

d1 = energy state of initial solution (measurementsequence)

d2 = energy state of neighborhood solution (new mea-surement sequence)

T = temperature

If acceptance criterion is satisfied, neighborhood solu-tion is accepted and stored otherwise control goes backand generates new neighborhood solution.

9. Process of generating neighborhood solution, its accep-tance or rejection has to be carried out for certain num-ber of times until following condition has been satisfied.

Fig. 8. Pseudo code for generating algorithm based on SA.

Fig. 7. Methodology to generate neighborhood solution.

S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91 85

Number of iterations (Niter) = maximum iterations(Maxiter).

In this way, pool of neighborhood solution has been cre-ated at a given temperature T. From this pool, measure-ment sequence with minimum distance (mind) andmeasurement sequence with maximum distance (maxd)has been selected to be used in later stages. This step helpsin great way by improving upon the neighborhood solutionand to get best possible solution.

10. Algorithm gets terminated with following condition:Ø < TS where, Ø ¼ ðmaxd�mindÞ

maxd and TS is stoppingtolerance.

If termination criterion is satisfied, measurement se-quence with minimum distance selected from the pool isaccepted as best measurement. Otherwise, temperature(T) has to be reduced depending on cooling factor andacceptance criterion has to be updated.

3.3. BF approach

BF approach is an exact method that generates andevaluates every possible solution. This approach is practi-cal only for relatively smaller problems and becomes inef-ficient as problem size increases. The approach thoughsimple and has wide applicability but is unacceptably slowfor large problems.

In present work, BF method has been used to validateresults obtained using algorithms based on GA and SA.The approach generates all possible measurement se-quences for given number of features and computes totaldistance moved in each of the measurement sequences.Different steps for BF method as presented in Fig. 9 canbe described as follows.

1. Identify number of features to be measured.2. Input coordinates for each measurement features.3. Generate all possible measurement sequences.

Fig. 9. BF method to obtain best possible measurement sequence.

86 S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91

4. Compute total distance moved (i.e. the cost) for eachgenerated measurement sequence.

5. Sort measurement sequences based on their mea-surement distances.

6. Choose measurement sequence with minimumdistance.

Pseudo code for BF based algorithm has been shown inFig. 10.

4. Implementation of algorithms

Application of proposed algorithms has been shownthrough a case study where disc of braking system (shownin Fig. 11) has to be inspected. This part has consisted ofnumber of features whose dimensional accuracies are veryimportant for their proper operation in assembly.

Inspection for this part requires measurement of manyfeatures at different positions. These features have to bemeasured one after another in particular measurement se-quence. Moreover, appropriate measurement sequenceresulting in reduced inspection time has been very impor-tant when large numbers of parts have to be inspected inmass production. Therefore, determination of appropriatemeasurement sequence from large number of possiblemeasurement sequences has been critical to overall perfor-mance of CMM inspection process. Hence, different algo-rithms have been implemented and compared to findmeasurement sequence which can provide minimum timefor given part. Ten measurement features identified for this

part have been designated as 1 for first feature, 2 for sec-ond feature and so on. These features included mainlyholes and planes. In Fig. 11b features 2, 3, 5, 6, 7 and fea-tures 4, 8, 10 represent point features on bottom and topplanes respectively while features 1 and 9 represent holeson top plane. Moreover, measurement sequence (6 4 10 2 97 5 8 1 3) indicate that CMM measurement probe shouldmeasure feature 6 first, then feature 4 and so onto finallyfinish measurement at feature 3.

1. Coordinates of features: First and foremost step of pres-ent algorithms requires determination of positions formeasuring features. Therefore, respective coordinatesof different features on given part have been identifiedand presented in Table 1.

2. Selection of parameters: There have been many param-eters to control the working of GA and SA based algo-rithms. In this work, preliminary studies and literatureinformation have been utilized to select parametersand corresponding levels. Parameters and different lev-els that have been identified for GA and SA based algo-rithms have been shown in Tables 2 and 3 respectively.

A series of experiments based on full factorial designhave been performed with the view to get appropriatecombination of levels for different parameters. Full facto-rial design resulted in 48 (3 � 42) and 81 (34) number ofexperiments for GA and SA based algorithms respectively.Moreover, five repetitions have been used for each experi-ment to avoid any random error. These algorithms havebeen executed for different parameter combinations usingMATLAB program on computer with Core2Quad 2.66 GHzCPU and 3.35 GB RAM. Outputs in terms of elapsed timefor program execution and distance obtained for measure-ment sequence have been collected. To get appropriatecombination of levels for different parameters, experi-ments have been analyzed for different outputs using Re-sponse Optimizer in statistical software. ResponseOptimizer determines combination of parameters settingsthat can collectively optimize set of responses. It providedan optimal solution for input variable combinations (suchas population size, crossover probability, mutation proba-bility and temperature, cooling factor, number of iterationsand stopping tolerance for GA and SA based algorithmsrespectively) using optimization plot. In this work, set ofresponses included elapsed time and total distance trav-elled in measurement sequence. Optimization plots forGA and SA based algorithms using response optimizercan be seen in Figs. 12 and 13 respectively.

Optimization plot resulted into following parametersettings for GA based algorithm:

� Population size = 8710.� Crossover probability = 0.80.� Mutation probability = 0.0050.

This parameter setting is expected to give measurementsequence with minimum distance in shortest possibleelapsed time for GA.

Optimization plot resulted into following parametersettings for SA based algorithm:

Fig. 10. Pseudo code for BF method.

Fig. 11. (a) Inspection process on CMM (b) Disc of braking system.

Table 1Coordinates of different features on part.

Features 1 2 3 4 5 6 7 8 9 10

Coordinates X 0 �77.9 65.4 37.5 �82.1 71 �101.4 �38.7 �49.8 �32.2Y �49.908 71.6 �74.3 36.8 �64.1 78.1 �5.5 �40.2 0.2 53.5Z �1.8069 �19.9 �19.8 0.1 �19.9 �19.9 �19.9 0.1 �2.9 0.1

S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91 87

Table 2Parameters and levels for GA based algorithms.

Parameters Numberof levels

Level values

1 2 3 4

Population size 3 100 1000 10000Crossover probability 4 0.65 0.75 0.85 0.95Mutation probability 4 0.005 0.01 0.05 0.1

Table 3Parameters and levels for SA based algorithms.

Parameters Numberof levels

Level values

1 2 3

Temperature 3 1,000,000 100,000 10,000Cooling factor 3 0.1 0.3 0.5Maximum number of

iterations3 1500 1000 500

Stopping tolerance 3 0.001 0.003 0.005

88 S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91

� Temperature = 10000.� Cooling factor = 0.46.� Maximum number of iterations = 1350.� Stopping tolerance = 0.0010.

This parameter setting is expected to give measurementsequence with minimum distance in shortest possibleelapsed time for SA based algorithm.

Since, BF is an exact method which identifies all possi-ble measurement sequences and computes all distances.Therefore, there was no need to determine parameter set-ting for its application. Moreover, results obtained with BFapproach (being most accurate method) have been used toconfirm results computed using GA and SA.

3. Execution of algorithms: Algorithms have finally beenimplemented with identified parameter settings inorder to compute best measurement sequence for given

Fig. 12. Optimization plot to obtai

part. Outputs of GA and SA based algorithms in terms ofmeasurement sequence, corresponding distance trav-elled and elapsed time have been presented in Table 4.

Results of GA and SA based algorithms have been con-firmed with results obtained with BF. Since, distancemoved obtained with GA and SA based algorithms is sameas that of BF approach. Therefore, it can be established thatmeasurement sequences (3 1 8 5 7 9 2 10 4 6) and (6 4 10 29 7 5 8 1 3) are best possible measurement sequences toinspect given part in minimum time.

4. Comparison of algorithms: Three approaches as pre-sented in Table 4 have been compared in terms of timetaken by them to obtain results. This comparison clearlysuggests that SA based algorithm has taken minimumtime of 2 s to provide measurement sequence for givenpart. It has also been noticed that BF approach has takenmaximum time of about 335 h 59 min 36.9 s. Moreover,time taken by BF approach increases with increase innumber of features. Therefore, effective application ofBF approach has been limited by maximum number offeatures.

GA based algorithm have taken almost 2 min 30.7 s toget best measurement sequence which is quite high incomparison to time taken by SA based algorithm. Thismay be due the fact that GA deals with large number ofindividuals in given population. Furthermore, minimumdistance of 541.926 mm for measurement sequences ob-tained using three approaches is same.

5. Results verification on CMM: Successful application ofabove mentioned approaches have been realizedthrough real time inspection of part on CMM. Generally,CMM programmers define measurement path formulti-feature component based on their experienceand expertise. This measurement path which do not

n parameter settings for GA.

Fig. 13. Optimization plot to obtain parameter settings for SA.

Table 4Results obtained after execution of algorithms.

Algorithms Measurementsequence

Distancemoved(mm)

Elapsed time

GA 3 1 8 5 7 9 2 10 4 6 541.926 2 min 30.7 sSA 6 4 10 2 9 7 5 8 1 3 541.926 2 s

3 1 8 5 7 9 2 10 4 6 541.926 2 sBF 6 4 10 2 9 7 5 8 1 3 541.926 335 h 59 min 36.9 s

S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91 89

follow any standard or systematic procedure have beendescribed here as randomly selected measurementsequence.

Table 5Inspection time of different measurement sequences on CMM.

Designation Measurement sequence Inspection time on CM

T1 T2

Identified measurement sequencesA 6 4 10 2 9 7 5 8 1 3 63.2 63.9B 3 1 8 5 7 9 2 10 4 6 64.0 63.2C 7 9 8 4 6 10 2 5 3 1 69.9 70.2D 5 8 3 1 9 7 10 2 4 6 66.5 66.6E 2 10 3 8 1 6 4 9 5 7 70.2 70.2F 6 3 9 1 8 5 7 2 10 4 68.8 67.6

Randomly selected measurement sequencesG 5 8 1 9 3 6 4 10 2 7 68.2 68.2H 2 7 5 8 1 3 9 4 6 10 68.5 68.1I 2 10 1 4 6 3 9 8 5 7 69.0 69.1J 5 1 9 3 4 6 10 2 8 7 71.1 71.0K 3 4 6 10 2 7 8 5 1 9 67.7 67.5L 8 1 9 3 4 6 10 2 7 5 68.5 68.7M 1 5 10 9 4 7 3 2 8 6 75.9 76.0N 8 9 10 2 3 1 6 4 5 7 71.7 71.6O 6 10 1 3 2 9 8 7 4 5 74.0 73.8P 9 3 4 2 10 1 5 6 7 8 75.2 75.1Q 9 1 7 2 6 3 5 8 10 4 72.8 72.7R 7 2 6 3 5 8 10 4 9 1 72.7 72.5

In this work given part has first been measured ran-domly using any measurement sequence and then throughidentified measurement sequence to compare performanceof different inspection processes. To maintain consistencyof measurement process, retract distance (=5 mm), clear-ance distance (=30 mm) and starting point (0,0,30) havebeen kept constant throughout all inspections.

Measurement sequences with corresponding designa-tion, mean inspection time and total measurement dis-tance can be seen in Table 5. It has been observed thatmeasurement sequences A and B have taken 541.926 mmof measurement distance in comparison to 1179.708 mmof distance taken by measurement sequence M. Therefore,decrease of almost 54% in distance moved by CMM probehas been observed with best measurement sequence.

M (s) Measurement distance (mm)

T3 Mean time (T)

64.0 63.7 541.92663.8 63.7 541.92670.0 70.0 778.50366.8 66.7 687.30370.2 70.2 801.03268.6 68.3 721.798

68.0 68.1 716.91667.3 68.0 705.46869.1 69.1 760.55571.1 71.1 822.79967.4 67.5 698.45468.3 68.5 725.23375.9 75.9 1179.70871.8 71.7 851.62373.7 73.8 992.18374.2 74.8 1098.27472.9 72.8 931.84871.3 72.2 914.391

90 S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91

Comparison of identified (A and B) and randomly se-lected measurement sequences (C to R) in Table 5 clearlysuggests minimum inspection time in case of measure-ment sequences A and B. This confirms that performanceof inspection process has improved with best measure-ment sequence. It has taken almost 63.7 s to completelymeasure the part using best measurement sequence (A).Moreover, time taken by measurement sequence (M) was75.9 s which was almost 19% greater than time taken bybest measurement sequence. In mass production, wherethousands of parts have to be inspected at a given workingshift, reduction of inspection time for overall improvementof inspection process becomes critical. For instance, mea-surement of 1000 parts would take almost 75920 s (21 h)if inspection is randomly carried out using measurementsequence M. Moreover, inspection with best identifiedmeasurement sequence (e.g., A in this case) would reduceinspection time to 63700 s (18 h). Therefore, almost 3 hof total production time can be saved with application ofpresent approaches.

5. Conclusions

In this paper, methodologies have been introduced tocompute sequence in which different part features shouldbe measured in minimum inspection time. Efficient appli-cation of these methodologies is very important for im-proved performance of CMM inspection process.

Three different approaches including BF, GA and SAbased algorithms have been utilized to identify best possi-ble measurement sequence for a given part. Best measure-ment sequence for measurement features results intominimum inspection time.

This research can be summarized as follows:

� SA based algorithm has taken minimum time of 2 s ascompared to BF approach which has taken maximumtime of 335 h 59 min 30.9 s. GA based algorithm havetaken almost 2 min 30.7 s to get best measurementsequence.� Exact method i.e., BF approach have resulted in

increased computing effort with increase in numberfeatures on the part. However, GA and SA based algo-rithms were fast and did not result in greater comput-ing effort with increased number of features.� These techniques are compact, robust and effective as

far as determination of measurement sequence for partinspection is concerned.� Application of present approaches can avoid repeated

movement of CMM probe on one path.� Execution of these methods suggests that given part

should be inspected in accordance to following mea-surement sequences.� 1 8 5 7 9 2 10 4 6) and (6 4 10 2 9 7 5 8 1 3).� These identified measurement sequences have resulted

in minimum distance of 541.926 mm.� 19% increase in inspection time has been observed

when part was measured using randomly selected mea-surement sequence M as compared to best measure-ment sequences (A and B).

� Implementation of such techniques is mandatory inmass production to minimize inspection time andhence overall production cost.� Present approaches to find appropriate measurement

sequence are practical and can easily be extended toobtain measurement sequence for inspecting any man-ufacturing part.

This work can be extended to include other perfor-mance measures of inspection process such as minimiza-tion of number of probe re-orientation which is relativelytime-consuming operation, reduction of stand-off distance(clearance, retract, approach distances etc.), strategiesselection etc. This work can also be elaborated to furtherimprove the working of GA and SA based algorithms byapplying new crossover operators, mutation operatorsand different neighborhood search method etc. The deter-mination of best measurement sequence using other meth-ods such as branch and bound, ACO can also be carried out.

Acknowledgment

Authors want to thank Advanced Manufacturing Insti-tute (AMI), King Saud University and Department of Indus-trial Engineering, King Saud University for their support.

References

[1] S. Topfer, G. Lin, U. Nehse, Inspection strategies and inspectionplanning for dimensional measurements of micro- and nano-structured components using cascaded sensor systems, JointInternational IMEKO TC1+TC7 symposium, Ilmenau, Germany, 2005.

[2] X. Ji, N. Li, ACO-Based Multiple Geometric Elements Measure PathPlanning for CMM, in: International Conference on ComputerApplication and System Modeling (ICCASM), China, 2010, pp. 539–544.

[3] Y. Lin, R. Mahabaleshwarkar, E. Massina, CAD based CMMdimensional inspection path planning—a generic algorithm,Robotica 19 (2001) 137–148.

[4] L. Qu, G. Xu, G. Wang, Optimization of the measuring path on acoordinate measuring machine using genetic algorithms,Measurement 23 (1998) 159–170.

[5] Z. Lin, C. Chen, Measuring-sequence planning by the nearestneighbor method and the refinement method, International Journalof Advance Manufacturing Technology 13 (1997) 271–281.

[6] E. Lu, J. Ni, S.M. Wu, An algorithm for the generation of an optimumCMM inspection path, Journal of Dynamic Systems Measurementand Control (Transactions of the ASME) 116 (1994) 396–404.

[7] R.O. Buchal, A. Wang, CMM sequence optimization with collisiondetection, International journal of Computer Applications inTechnology 26 (2006) 65–74.

[8] S.M. Ruegsegger, Intelligent scheduling optimization using a rule-based artificial neural network, Aerospace and ElectronicsConference, NAECON, Proceedings of the IEEE 2 (1993) 800–806.

[9] C.G. Lu, D. Morton, P. Myler, M.H. Wu, An artificial intelligent (AI)inspection path management for multiple task measurement on co-ordinate measuring machine (CMM): an application of neuralnetwork technology, in: Proceedings of 31st internationalMATADOR conference, 1995, pp. 353–357.

[10] F. Cus, J. Balic, Optimization of cutting process by GA approach,Robotics and Computer-Integrated Manufacturing 19 (2003) 113–121.

[11] F. Kolahan, A.H. Khajavi, Modeling and optimization of abrasivewater jet parameters using regression analysis, World Academy ofScience, Engineering and Technology 35 (2009) 488–493.

[12] N. Kaya, Machining fixture locating and clamping positionoptimization using genetic algorithms, Computers in Industry 57(2006) 112–120.

[13] M. Huang, C. Hsieh, J.S. Arora, A genetic algorithm for sequencingtype problems in engineering design, International Journal forNumerical Methods in Engineering 40 (1997) 3105–3115.

S.H. Mian, A. Al-Ahmari / Measurement 47 (2014) 78–91 91

[14] J.A Qudeiri, H. Yamamoto, R. Ramli, Optimization of operationsequence in CNC machine tools using genetic algorithm, Journal ofAdvanced Mechanical Design, Systems, and Manufacturing 1 (2007)272–282.

[15] C.Y. Low, K.H. Chong, K. Salleh, K.S.P. Johnny, Path optimizationusing genetic algorithm evolution, in: Proceedings of IEEEStudent Conference on Research and, Development, 2010,pp. 252–255.

[16] D.P. Garg, M. Kumar, Optimization techniques applied to multiplemanipulators for path planning and torque minimization,Engineering Applications of Artificial Intelligence 15 (2002) 241–252.

[17] X. Geng, Z. Chen, W. Yang, D. Shi, K. Zhao, Solving the travelingsalesman problem based on an adaptive simulated annealingalgorithm with greedy search, Applied Soft Computing 11 (2011)3680–3689.

[18] R. Baraglia, J.I. Hidalgo, R. Perego, A hybrid heuristics for thetravelling salesman problem, IEEE Transactions on EvolutionaryComputation 5 (2001) 613–622.

[19] S.M. Chen, C.Y. Chien, Solving the travelling salesman problem basedon the genetic simulated annealing ant colony system with particleswarm optimization techniques, Expert Systems with Applications38 (2011) 14439–14450.

[20] W. Zicheng, G. Xiutang, S. Zehui, An effective simulated annealingalgorithm for solving the travelling salesman problem, Journal ofComputational and Theoretical Nanoscience 6 (2009) 1680–1686.

[21] E. Kodeekha, Brute force method for lot streaming in FMS schedulingproblems, intelligent engineering systems, in: IEEE InternationalConference, Budapest, 2007, pp. 179–184.

[22] C. Chauhan, R. Gupta, K. Pathak, Survey of methods of solving TSPalong with its implementation using dynamic programmingapproach, International Journal of Computer Applications 52(2012) 12–19.

[23] J.E. Baker, Adaptive selection methods for Genetic algorithms, in:Proceedings of the 1st International Conference on Genetic Algorithmsand their Applications, Pittsburgh, USA, 1985, pp. 101–111.

[24] L. Davis, Applying adaptive algorithms to epistactic domains, in:Proceedings of the International Joint Conference on ArtificialIntelligence – IJCAI, 1985, pp. 162–164.

[25] L. Davis, Handbook of Genetic Algorithms, Van Nostrand Reinhold,New York, 1991.

[26] W. Banzhaf, The ‘‘molecular’’ travelling salesman, BiologicalCybernetics 64 (1990) 7–14.

[27] R.M. Selvi, R. Rajaram, Effect of crossover operator in geneticalgorithms on anticipatory scheduling, in: 24th Internationalsymposium on automation and robotics in construction (ISARC),2007, pp. 369–373.

[28] D.B. Fogel, L.C. Stayton, On the effectiveness of crossover insimulated evolutionary optimization, Biosystems 32 (1994) 171–182.

[29] R.W. Eglese, Simulated annealing: a tool for operational research,European Journal of Operational Research 46 (1990) 271–281.