Available online at www.sciencedirect.com
ScienceDirect Procedia CIRP 57 (2016) 504 – 509
49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016)
Static versus dynamic provision of worker information in manual assembly: a comparative study using eye tracking to investigate the impact on productivity and added value based on industrial case examples Mario Lušiüab*, Christian Fischerb, Konrad Schmutzer Braza, Marina Alamb, Rüdiger Hornfecka, Jörg Frankeb b
a Technische Hochschule Nürnberg, Institute for Chemistry, Materials and Product Development, Keßlerplatz 12, 90489 Nürnberg, Germany Friedrich-Alexander-University Erlangen-Nuremberg / Institute for Factory Automation and Production Systems, Egerlandstraße 7, 91058 Erlangen, Germany
* Corresponding author. Tel.: +49-911-5880-1906; fax: +49-911-5880-5900. E-mail address:
[email protected]
Abstract Manual product assembly is usually done by providing worker information accordingly to the present assembly task. Traditionally, paper-based systems are used for this purpose, but nowadays, IT-based worker information is gaining ground. Those systems in particular allow the transmission of dynamic information and, thus, enable a realistic and timely representation of assembly sequences. Previous comparative studies on the effective difference between static and dynamic provision of worker information have two gaps. On the one hand, the majority investigated less manual assembly tasks close to real industrial cases. The experiments take place rather on abstracted assembly models under research laboratory conditions. On the other hand, there is a lack of studies on the efficiency of the interaction between the information providing medium, the available assembly objects and the actual worker’s behaviour. This is regarded in a way simultaneous to the assembly process and from a real view of the worker. To be able to draw conclusions about the efficiency of this interaction, a real view registry, which is provided by the yet sparingly applied eye tracking technology, is feasible. The experiments described in this article draw on both aspects to realise a lifelike as possible comparison between static and dynamic provision of worker information: Based on industrial case examples, and thus close to a real working environment, the worker’s eye movement is documented specifically for the assembly tasks and afterwards holistically analysed in concern of the impact on productivity and value adding. © Authors. Published by Elsevier B.V. This ©2016 2015The The Authors. Published by Elsevier B.V.is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of Scientific committee of the 49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016). Peer-review under responsibility of the scientific committee of the 49th CIRP Conference on Manufacturing Systems Keywords: Assembly; Information; Productivity; Value adding; Eye tracking.
1. Introduction
2. Need for action
Assembly information support improves the working performance of a worker [1]. Traditionally, paper-based systems are used for this purpose [2], recently, however, ITbased worker information systems are to be found [3,4]. These in particular allow the transmission of dynamic information and thus enable realistic representation of assembly sequences [5]. As addressed in the following chapter, previous comparative studies on the effective difference between static and dynamic provision of worker information have two gaps.
There is a bunch of publications differentiating static and dynamic representation of information. These publications are previously filtered by some elimination criteria that are described in the following paragraphs. One criterion is the kind of knowledge that is actually requested by a worker. As this study only concerns the usage of an information system in a real assembly environment, we can exclude publications based on declarative knowledge. Examples for the latter are academic topics like photosynthesis [6], the functionality of a cable pulley system [7] or first aid theory [8]. The mentioned publications
2212-8271 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 49th CIRP Conference on Manufacturing Systems doi:10.1016/j.procir.2016.11.087
505
Mario Lušić et al. / Procedia CIRP 57 (2016) 504 – 509
predominantly concern the test of knowledge and comprehension, only within a few experimental series, actual assembly tasks are included. However, most experimental subjects were pupils that had to knot [9] or to fold paper airplanes [10]. To get a relation to manual assembly processes in an industrial context, we choose publications with realistic assembly tasks. We go on analysing publications which deal with the subjects’ motor skills as well as problem-solving and procedure-oriented methods. This paper focusses industrial assembly processes with real parts from automotive production, which is one gap identified within the concerned pertinent research. The used working instructions base on three-dimensional (3-D) computer-aided design (CAD) models. For worker guidance either screenshots or animated 3-D models, that show the assembly paths of parts, were used. An important differentiator for literature analysis is the role of animations used within experiments. They can be merely decorative and, thus, may contribute to a subject’s motivation or attention, or they indeed carry an informational value [11]. In addition, according to that meta-analysis of Höffler and Leutner, that focuses predominantly on declarative than procedural knowledge, representing animations are more effective than representing static pictures. However they don’t identify significant differences between decorative animations and static pictures [11]. Consequently, we only consider publications with representing animations. A further publication about the effect of different kinds of signals, like coloured high-lighting on an experimentee [12], can be excluded although eye-tracking has been used, hence the dynamic in information provision only means signal change. Table 1 gives an overview about the selected publications for further consideration after our analysis.
[9] – exp. 1
9
9
[9] – exp. 2
Duration of initial reading in
Recognition of previous/subsequent steps
Frequency of look up and required look up time
Number of attempts
Number of correctly done working steps
Number of errors
Working period
Publication
Table 1. Comparison of relevant papers and their experimental parameters
9
9
[13]
9
[14]
9
[15]
9
[16]
9
9
[17]
9
9
[18]
(9)
[19]
9
[20]
9
[21]
9
9
9
9
9
9
9
[10] – exp. 2
9
[10] – exp. 3
9
3.1. Dependent variables Subsequent to the findings of chapter 2 the following dependent variables were measured: working duration, time for additional look up at instruction, frequency of additional look up at instruction and number of properly performed assembly steps. Working duration and time for additional look up at instruction. The working duration is basically divided into assembly time and look up at instruction time. During the assembly time the worker is executing value-adding activities, such as assembling parts or using tools. The additional look up at instruction time means, in contrast, not value-adding periods, since the worker interrupts the assembly process for reading or looking on the worker information system. The additional look up, due to the repeated understanding of the assembly task, occurs after the initial read in time, which is not limited but determined by the worker himself, if he decides to start the assembly. Fig. 1 illustrates the structuring of assembly time existent from once initial read in time, multiple value-adding assembly periods and multiple periods for additional look up at instruction. step a i. r. i.
l. u.
a
9
working duration a = time for assembly ؙvalue adding time i. r. i. = time for initial read in l. u. = time for additional look up at instruction
9 9
Following an approach close to industrial reality, the present study focuses on variables that are essential for pertinent categories like productivity and added value. The crucial elements of the experimental setup that is methodologically based on the conceptualisation of the psychological experiment by Sarris [24], are described below.
9
9 9
3. Experimental setup
9 9
[10] – exp. 1
studies it is involved to track, if a subject notices certain signalling techniques [12], but its potential for a documentation of the actual behaviour of workers has not yet been exploited. In this context, one should have a look at processing time as an important parameter, especially for assembly processes. The processing time is a sum of the working time for actual activity on parts and of the look up time for reviewing working instructions, which can be determined as a no-valueadding activity. Eye tracking technology enables the measurement and the evaluation of periods in an easy, detailed, and reproducible way. The real watching time on distinct areas of interest [23] is recorded and can be redrawn later for evaluation. Certain areas of interest might be working instruction on screens or assembly environments.
a
l. u.
a
a
l. u.
a
step a+1 i. r. i. a
9 Fig. 1. Structuring of time required for each assembly step 9
Another gap concerning the pertinent research is the scarce application of eye tracking technology [22] so far. In some
Frequency of additional look up at instruction. The rate of the participants’ looks ups back at the instruction is
506
Mario Lušić et al. / Procedia CIRP 57 (2016) 504 – 509
measured. E.g. in Fig. 1, the frequency of additional look up at instruction during performing the assembly step is at three times. This frequency is assumed as an indicator for the quality of the explanation of the assembly task by. The more the worker looks back at the instruction, the worse one could rate the task description provided by the medium that is delivering information. Number of properly performed assembly steps. For each assembly step properly performed there is one point. In both experiments, a maximum of 15 points can be reached. To avoid subsequent errors, assembly errors were corrected by the supervisor before the next assembly step is started. Meanwhile, time tracking is stopped. 3.2. Hypotheses
cited in chapter two were compared and examined for similarities (Fig. 2). The characteristics that were actually applied to design the assembly instructions within this study are highlighted in grey in Fig. 2. design criteria Type of presentation
drawing
photo
Static solution: displaying individual frames
sequential
Static solution: degree of interactivity of the participant
none
Dynamic solution: interactivity of the participant
The investigation is guided by the following hypotheses: Language setting
x Providing worker information dynamically, in comparison with the provision of static worker information, results in a faster completion of the assembly task (hypothesis 1). This would result in an increased productivity. x Providing worker information dynamically, in comparison with the provision of static worker information, results in a decrease in time for and frequency of additional look up at instruction (hypothesis 2). This would be tantamount to a minimisation of the proportion of non-value-adding activities within assembling. x Providing worker information dynamically results, in comparison with the provision of static worker information, in a lower number of assembly errors (hypothesis 3). x The effect assumed within hypotheses 1, 2 and 3 should be more pronounced with a higher complexity of the assembly task (hypothesis 4). This expectation corresponds to the assumption made in [14], arguing that the effect of faciliation by reducing cognitive load should increase with the complexity of the assembly task.
manifestation static
Using signalling
none
dynamic animation
video
semi-simultaneously
simultaneously
any go back and forth
Restricted play, pause, to play fast forward, rewind and pause
additionally: setting timelapse, slowmotion
with text
without text
with Audio
without Audio
none
arrows
colour highlighting
hands
Fig. 2. Common assembly instruction design criteria and expressions found within the publications considered in chapter 2.
On each slide (Fig. 3) a short text names the current assembly step. With a dynamic information provision, a video is played, while, in case of a static provision, two pictures connected by an arrow are displayed to represent the initial state before and the final state after assembling.
3.3. Assembly objects based on industrial case examples In experiment I a subassembly of a throttle body had to be assembled, while in experiment II a subassembly of an air nozzle was used. The throttle body consists of 19 parts, the air nozzle covers 14 parts but both had to be assembled in 15 steps. The assembly of the air nozzle is considered as less complicated than the throttle body, hence the air nozzle is assembled only by multiple shifting one into another, but the throttle body applies additionally with the operation of screwing. In addition, for the assembly of the air nozzle, no further tools are necessary, but for the throttle body tools like pliers and various screwdrivers are required. Moreover, generally the parts of the throttle body are smaller and not as easily distinguishable as the parts of the air nozzle.
Fig. 3. Information providing medium: dynamic (a) versus static (b) design of assembly instructions used within this investigation1.
3.5. Documentation of worker’s behaviour using eye tracking To be able to draw conclusions on the efficiency of the interaction between the information providing medium, the
3.4. Independent variables For the selection of the independent variables respectively the design of common assembly instructions, the publications
1 Created by using Dassault Systems’ software 3DVIA Composer and embedding therefrom resulting files into PowerPoint presentations.
507
Mario Lušić et al. / Procedia CIRP 57 (2016) 504 – 509
available assembly objects and the actual worker’s behaviour, an implementation of the eye tracking technology is valuable, hence the real view of the worker can be captured simultaneously to the assembly process. The presently most widespread technique for eye tracking is the method of corneal reflection. We therefore used Tobii glasses, worn by a test subject (Fig. 4).
seconds
Working duration 500 450 400 350 300 250 200 150 dynamic
static
Fig. 5. Boxplot working duration, experiment 1
Time for additional look up at instruction (Fig. 6). A difference can be seen also in the time for additional look up at instruction. The additional look up time during the use of dynamically provided worker information was in average 25sec. (median 20sec.; standard deviation = 12sec.). The static group needed in average 39sec. (median 37sec.; standard deviation = 9sec.). Fig. 4. An experimentee wearing Tobii Glasses while assembling Time for additional look up at instruction
3.6. Experimental procedure 60
4. Results Subsequently, the results for the experiments 1 and 2 are listed.
50 seconds
There were two different assembly experiments (static versus dynamic) each performed by 22 participants. The testpersons had been instructed to carry out the assemblies stepwise in accordance with the information provided by the worker information system. The assembly objects were placed in front of the monitor on the assembly bench and named with a label. In addition, all required tools were provided. While regarding the dynamic information provision, the participant had the possibility to pause or repeat each video respectively each assembly step as required. The assembly time was not limited. If an assembly step was performed wrong, the participant was informed and the assembly mistake was been corrected in order to avoid subsequent errors. After the completion of 15 predefined assembly steps the experiment ends.
40 30 20 10 dynamic
static
Fig. 6. Boxplot time for additional look up, experiment 1
Frequency of additional look up at instruction (Fig. 7). The frequency of additional look up was smaller while providing worker information in a dynamic way, too, since it reached an average of 20 look ups (median = 19 times; standard deviation = 6 times). In comparison, the group using static instructions had to acquire repeated information 31 times in average (median = 32 times; standard deviation = 7 times). Frequency of additional look up at instruction
4.1. Experiment 1 – assembling throttle body Working duration (Fig. 5). Using a dynamic provision of worker information the working duration was in average at 257seconds (median = 263sec.; = standard deviation = 47sec.). The static group took in average 51sec. longer for the same task (median = 302sec.; standard deviation = 60sec.).
50 40 30 20 10 dynamic
static
Fig. 7: Boxplot frequency of additional look up at instruction, experiment 1
Number of properly performed assembly steps. With an average number of properly assembly steps of 14.8 out of 15
508
Mario Lušić et al. / Procedia CIRP 57 (2016) 504 – 509
(standard deviation = 0.57), slightly less errors occurred within the group of static information provision compared to the group using dynamic information provision (average = 14.4; standard deviation = 0.88).
Frequency of additional look up at instruction 50 40
4.2. Experiment 2 – assembling air nozzle
30
Working duration (Fig. 8). Regarding the condition of dynamic provision of worker information the working duration reached an average of 291sec. (median = 286sec.; standard deviation = 57sec.). The static group needed 304sec. in average (median = 279sec.; standard deviation = 72sec.).
20
seconds
dynamic
static
Fig. 10. Boxplot frequency of additional look up at instruction, experiment 2
Number of properly performed assembly steps. The average number of correctly performed assembly steps is at 14.9 (standard deviation = 0.3) for dynamic information provision and reaches 14.3 (standard deviation = 0.6) for static information provision.
Working duration 500 450 400 350 300 250 200 150
5. Summary, discussion and future work dynamic
static
Fig. 8. Boxplot working duration, experiment 2
Time for additional look up at instruction (Fig. 9). Focusing on this variable a difference reveals, too. The additional look up time during the use of dynamically provided worker information was in average 23sec. (median 23sec.; standard deviation = 7sec.). The static group needed in average 38sec. (median 34sec.; standard deviation = 10sec.). Time for additional look up at instruction 60 50 seconds
10
40 30 20 10 dynamic
static
Fig. 9. Boxplot time for additional look up, experiment 2
Frequency of additional look up at instruction (Fig. 10). The frequency of additional look up was smaller while providing worker information in a dynamic way, too, since it reached an average of 20 look ups (median = 17 times; standard deviation = 5 times). In comparison, the group using static instructions had to acquire repeated information 28 times in average (median = 26 times standard deviation = 6 times).
We compared the influence of static und dynamic provision of worker information on the efficiency of manual assembling. Therefore test items that are as close to the industrial reality of manual assembling as possible were applied. For our experiment we focused independent variables that are relevant for the characterisation of productivity and value-adding. To be able to draw conclusions on the efficiency of the interaction between the information providing medium, the available assembly objects and the actual worker’s behaviour, the eye tracking technology was used to capture the real view of the worker simultaneously to the assembly process. Hypothesis 1 can be confirmed, since regarding both experiments, the groups that had been using dynamic assembly instructions completed the tasks faster. As a consequence a higher flow rate in manual assembling becomes possible. Focusing the proportional distribution of the working time, hypothesis 2 is confirmed, too. While using dynamic visualisations for information provision the experimental subjects need less reassurance about the assembling procedure, since the frequency of additional lookup decreases in comparison with a static information provision. Thereby, covering both experiments, the dynamic information delivery reduces the time for additional look up at instruction. Consequently, the non-value-adding parts of the assembling work can be minimized. Regarding the number of errors, no significant difference appears between the two principles of worker information delivery. In consideration of both experiments hypothesis 3 could not be confirmed. However, the use of dynamic provision of worker information results in faster processing with a constant quality of task handling, and therefore conduces a higher productivity in manual assembly work. Hypothesis 4 can mainly be confirmed, since the effect of time saving increases with the more complex assembly of experiment 1. Solely for the time of additional look up the measured results are virtually the same. Additionally, the
Mario Lušić et al. / Procedia CIRP 57 (2016) 504 – 509
findings concerning hypothesis 2 suppose that a dynamic and representationally denser information provision facilitates manual assembling, especially with complex assembly tasks. For a further exploration of the nature of this issue, the investigation of the subjective task-comprehension would be valuable. Therefore, also some qualitative research seems appropriate. In addition, the perceived task complexity should be measured by adapting a validated item, e.g. the NASATask Load Index [25]. To affirm the findings, further research should involve more test-persons and a higher number of diverse assembling objects. The most desirable test-condition would, of course, be a real industrial surrounding with workers doing their daily manual assembly work. As the evaluation of the efficiency of worker information systems by eye tracking technology seems to be a promising and feasible method, a short, practical and industry-near solution should be found. Thus, a company could analyse each work place under consideration of a worker’s concentration on assembly processes and information reception. This contributes to an optimisation of processes for working places’ efficiency and ergonomics and also shop floor. Acknowledgment This work was supported by Boryszew Kunststofftechnik Deutschland GmbH (Magdeburg, Germany) and BING Power Systems GmbH (Nürnberg, Germany) who provided industrial subassemblies from real serial production. This gave us the opportunity to create an analysis as close as possible to the real tasks during manual assembly. References [1] Tan J, Duan F, Zhang Y, Watanabe K, Pongthanya N, Sugi M, Yokoi H, Arai T. Assembly Information System for Operational Support in Cell Production. In: Mitsuishi M, Ueda K, Kimura F, editors. Manufacturing Systems and Technologies for the New Frontier. London: Springer; 2008. p. 209–212. [2] Roller D. Technical Information System for Assembly, Test and Service Documentation. In: Brödner P, Karwowski W, editors. Ergonomics of Hybrid Automated Systems III. Proceedings of the Third International Conference on Human Aspects of Advanced Manufacturing and Hybrid Automation, Gemany, August 26-28, 1992. Amsterdam: Elsevier Science Publishers B.V.; 1992. p. 273–278. [3] Lušiü M, Hornfeck R, Fischer C, Franke J. Lean Information Management of Manual Assembly Processes: Creating IT-Based Information Systems for Assembly Staff Simultaneous to the Product Engineering Process. Applied Mechanics and Materials 2013; 421. p. 546–553. [4] Lušiü M, Fischer C, Bönig J, Hornfeck R, Franke J. Worker information systems: state of the art and guideline for selection under consideration of company specific boundary conditions. Procedia CIRP 2016; 41. p. 1113–1118. [5] Lušiü M, Schmutzer Braz K, Wittmann S, Fischer C, Hornfeck R,
Franke J. Worker Information Systems including Dynamic Visualisation: a Perspective for Minimising the Conflict of Objectives between a Resource-efficient Use of Inspection Equipment and the Cognitive Load of the Worker. Advanced Materials Research 2014; 1018. p. 23–30. [6] Höffler TN, Prechtl H, Nerdel C. The influence of visual cognitive style when learning from instructional animations and static pictures. Learning and Individual Differences 2010; 20-5. p. 479–483. [7] Boucheix JM, Schneider E. Static and animated presentations in learning dynamic mechanical systems. Learning and Instruction 2009; 19-2. p. 112–127. [8] Arguel A, Jamet E. Using video and static pictures to improve learning of procedural contents. Computers in Human Behavior 2009; 25-2. p. 354–359. [9] Ayres P, Marcus N, Chan C, Qian N. Learning hand manipulative tasks: When instructional animations are superior to equivalent static representations. Computers in Human Behavior 2009; 25-2. p. 348–353. [10] Wong A, Marcus N, Ayres P, Smith L, Cooper GA, Paas F, Sweller J. Instructional animations can be superior to statics when learning human motor skills. Computers in Human Behavior 2009; 25-2. p. 339–347. [11] Höffler TN, Leutner D. Instructional animation versus static pictures: A meta-analysis. Learning and Instruction 2007; 17-6. p. 722–738. [12] Boucheix JM, Lowe RK, Putri DK, Groff J. Cueing animations: Dynamic signaling aids information extraction and comprehension. Learning and Instruction 2013; 25. p. 71–84. [13] Castro-Alonso JC, Ayres P, Paas F. Animations showing Lego manipulative tasks: Three potential moderators of effectiveness. In: Computers & Education 2015; 85. p. 1–13. [14] Garland TB, Sanchez CA. Rotational perspective and learning procedural tasks from dynamic media. Computers & Education 2013; 69. p. 31–37. [15] Lee DY, Shin DH. An empirical evaluation of multi-media based learning of a procedural task. Computers in Human Behavior 2012; 28-3. p. 1072–1081. [16] Michas IC, Berry DC. Learning a procedural task: effectiveness of multimedia presentations. Appl. Cognit. Psychol. 2000; 14-6. p. 555– 575. [17] Palmiter S, Elkerton J. Animated Demonstrations for Learning Procedural Computer-Based Tasks. Human–Computer Interaction 1993; 8-3. p. 193–216. [18] Schwan S, Riempp R. The cognitive benefits of interactive videos: learning to tie nautical knots. Learning and Instruction 2004; 14-3. p. 293–305. [19] Spangenberg RW. The motion variable in procedural learning. AV Communication Review 1973; 21-4. p. 419–436. [20] van Hooijdonk C, Krahmer E. Information modalities for procedural instructions: The influence of text, pictures, and film clips on learning and executing RSI exercises. IEEE Transactions on Professional Communication 2008; 51-1. p. 50–62. [21] Watson G, Butterfield J, Curran R, Craig C. Do dynamic work instructions provide an advantage over static instructions in a small scale assembly task? Learning and Instruction 2010; 20-1. p. 84–93. [22] Bergstrom JR, Schall AJ. Eye tracking in user experience design. Amsterdam a.o.: Elsevier, Morgan Kaufmann; 2014. [23] Holmqvist K, Nyström M, Andersson R, Dewhurst R, Jarodzka H, Van de Weijer J. Eye tracking. A comprehensive guide to methods and measures. Oxford: Oxford University Press; 2011. [24] Sarris V. Methodologische Grundlagen der Experimentalpsychologie. Band 1. Erkenntnisgewinnung und Methodik der experimentellen Psychologie. Munich a. o.: Ernst Reinhardt, UTB; 1990. [25] Hart S, Staveland L. Development of NASA-TLX (Task Load Index): Results of empirical and theoretical research. In: Hancock PA, Meshkati N, editors. Human Mental Workload. Amsterdam a.o.: NORTHHOLLAND; 1988. p. 139–184.
509