A comparative study of assembly planning in traditional ... - IEEE Xplore

2 downloads 104 Views 1MB Size Report
environments over the traditional engineering environment in improving the subjects' overall assembly planning performance and in minimizing the handling ...
546

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 29, NO. 4, NOVEMBER 1999

A Comparative Study of Assembly Planning in Traditional and Virtual Environments Nong Ye, Prashant Banerjee, Amarnath Banerjee, and Fred Dech Abstract—This paper presents an experiment that investigated the potential benefits of virtual reality (VR) environments in supporting assembly planning. In the experiment, fifteen subjects performed an assembly planning task in three different conditions: a traditional engineering environment (TE), a nonimmersive desktop VR (DVR) environment, and an immersive care automatic virtual environment (CAVE) VR environment (CVR environment). The effects of the three conditions on the subjects’ performance were analyzed. The subjects’ performance time in the TE condition was significantly longer than that in the DVR condition and that in the CVR condition, whereas the difference in performance time between the DVR condition and the CVR condition was not significant. The total number of problematic assembly steps in the TE condition was significantly greater than that in the CVR condition. Specifically, the subjects’ assembly sequences in the TE condition involved more reorientations than in the DVR condition. The number of difficult assembly steps in the TE condition was significantly greater than that in the DVR condition, which was significantly greater than that in the CVR condition. The number of dissimilar assembly steps in the TE condition was significantly greater than that in the CVR condition, which was significantly greater than that in the DVR condition. Hence, the results revealed advantages of the two VR environments over the traditional engineering environment in improving the subjects’ overall assembly planning performance and in minimizing the handling difficulty, excessive reorientation, and dissimilarity of assembly operations. Index Terms— Assembly planning, computer-aided manufacturing, rapid phototyping, virtual reality

I. INTRODUCTION

A

SSEMBLY planning determines the sequence and process details of assembly operations that put individual parts together into an assembly [1]–[5]. An assembly plan has a large impact on production efficiency and costs. Assembly planning usually is performed by production engineers for an assembly design. Since assemblies have become increasingly complex (i.e., involving hundreds of parts), assembly planning has presented a considerable challenge to production engineers. Many factors must be considered in assembly planning [1]–[6]. For example, production engineers must examine the geometric design of an assembly to ensure a feasible assembly sequence that does not induce part collisions and part trappings. Production engineers also need to look into other factors such as the reorientation, directionality, stability, manipulability, and parallelism of assembly operations, as well as the complexity of tools and fixtures. Those factors Manuscript received March 28, 1998; revised March 18, 1999. N. Ye is with the Department of Industrial Engineering, Arizona State University, Tempe, AZ 85287-5906 USA. P. Banerjee, A. Banerjee, and F. Dech are with the Department of Mechanical Engineering, University of Illinois, Chicago, IL 60607 USA. Publisher Item Identifier S 1094-6977(99)08206-1.

determine a good assembly sequence with respect to efficiency, costs, product safety, and operator safety relating to assembly operations [6]–[12]. Aiming at easing the assembly planning task for production engineers, many studies have been carried out to automate the generation of assembly plans [13]–[28]. Since the assembly sequence is the backbone of an assembly plan, most of the efforts have focused on the automatic generation of assembly sequences. Although progress has been made to generate feasible assembly sequences even for complex assemblies, there still exist difficulties in generating good assembly sequences [6], [11]. First of all, it is difficult to quantify goodness criteria for computerizing the goodness evaluation of assembly sequences, not to mention that there may exist conflicts among goodness criteria. Another difficulty lies in the computational complexity of searching for good assembly sequences in a large search space of feasible assembly sequences. The presentation of feasible assembly sequences to production engineers for a manual selection of good assembly sequences is not practical, because a large number of unfamiliar assembly sequences simply confuse and overwhelm production engineers. Since there is a long way to realize the automatic generation of assembly sequences, assembly planning still relies on production engineers. Caldwell, Ye, and Urzi reported a study with a manufacturing corporation to analyze computer tools available to support assembly planning [29]. It was concluded that many computer-aided design/computer-aided manufacturing (CAD/CAM) systems used by production engineers provided little support to assembly planning. Decisions for transforming an assembly design into an assembly plan and then to an assembly production system were made manually and were communicated between departments (e.g., product design, production engineering, and manufacturing facility, which were often geographically separated) mostly through conventional channels rather than the electronic channel of information sharing. Recently, ideas of supporting the assembly planning task of production engineers in ways other than the automatic generation of assembly sequences have emerged [11], [30]–[34]. Among those ideas, assembly planning in a virtual reality (VR) environment has attracted much attention [35]–[36]. Advances in automatically loading CAD data of an assembly design into a VR environment have enabled the rapid prototyping of an assembly and its parts in a VR environment [31]–[32]. This has opened a new avenue for assisting production engineers in assembly planning. Through viewing and manipulating a virtual assembly and virtual parts, production engineers are able to

1094–6977/99$10.00  1999 IEEE

YE et al.: A COMPARATIVE STUDY OF ASSEMBLY PLANNING

create, express, simulate, and evaluate an assembly sequence in a virtually natural but more cost-effective setting. Production engineers are able to examine various aspects (i.e., feasibility and goodness) of an assembly sequence by performing virtualassembly operations on virtual parts just as real parts are being put together through real assembly operations into a real assembly. Thus, the assembly sequence can be tested while it is generated in a VR environment. Furthermore, the simulation of assembly operations can be saved, demonstrated, and printed out as a specification of the assembly plan. Therefore, a VR environment will enable an integration of assembly design and assembly planning, a more realistic presentation of an assembly design, a simulation-based testing and verification of assembly planning outcome, and an integration of assembly planning and assembly specification. In contrast, in a traditional engineering environment, assembly design and assembly planning are performed separately by people in different departments. After product designers complete an assembly design and generate product drawings, production engineers perceive an assembly through its design drawings and generate an assembly sequence. Since both the understanding of an assembly and the generation of an assembly sequence are performed mentally, assembly planning is a cognitively demanding and thus error-prone task. Production engineers often have to use physical prototypes of real parts and trials of real production runs for verifying their assembly plan and revealing any errors. Prototypes of real parts and trials of real production runs come with high costs and prolonged cycles of assembly planning. Therefore, a VR environment has the potential to offer a more natural, powerful, economic, flexible platform than a traditional engineering environment to support assembly planning. This paper presents an experiment that examines the potential benefits of using VR environments to support assembly planning by comparing the assembly-planning performance of subjects in traditional and VR environments. II. EXPERIMENT A. Apparatus In the experiment, two VR environments were investigated: a nonimmersive desktop VR environment (DVR) and an immersive CAVE VR environment (CVR). The DVR environment that we used for our experiment consisted of a Silicon Graphics workstation with IRIX CosmoPlayer VRML 2.0 browser plug-in to Netscape [37]. The CVR environment uses an IRIS Performer CAVE interface developed at the Electronic Visualization Laboratory, University of Illinois, Chicago [38]. The care automated virtual environment (CAVE) is a 10 10 9-foot room that uses rear-projected, high-resolution projectors to produce an immersive, three-dimensional (3-D) environment. The commercially available CAVE environment produces a 3-D stereo effect by displaying, in alternating succession, the left and the right-eye views of the scene as rendered from the viewer’s perspective. These views are then seen by the user through a pair of LCD shutter glasses whose lens opens and closes 48 times

547

per second in synchronization with the left and right-eye views. The correct viewer-centered projection is calculated based on the viewer’s position and orientation as determined by an electromagnetic tracking system. The position and orientation of a 3-D wand are also tracked. This wand allows for navigation in the virtual world. The immersion in CVR provides subjects with a more realistic sense of virtual assemblies and parts. However, CVR costs much more than DVR. It is likely that a manufacturing company is willing to set up only one fully immersive VR environment and that the access to the VR environment is restrictive. A low-end VR display on a desktop computer, such as the DVR environment in this study, is more affordable and accessible, although it is not clear how well it supports assembly planning in comparison to a high-end, immersive VR environment. Therefore, in this study we investigated both a high-end immersive VR environment and a low-end, nonimmersive VR environment to examine possible performance differences between them.

B. Design of Experiment The experiment was based on a between-subject factorial design. The assembly-planning environment as the independent variable has three conditions: a traditional engineering environment (TE), a nonimmersive DVR environment, and an immersive CVR environment. The TE condition provided the basis of comparison with the VR conditions. The three conditions differed in ways in which an assembly was presented and manipulated. The assembly used in the experiment was an air-cylinder that consisted of 34 parts. The assembly was a real product from a company. A main reason for selecting this assembly for the study is the moderate complexity of the assembly (i.e., the number of parts). The assembly presented a certain level of task difficulty to the subjects, and yet allowed them to complete the experiment within a reasonable period of time without fatigue. For the TE condition, the air-cylinder assembly and each of its parts were presented using a stack of hard-copy drawings. The drawings were generated using the commercial package ProEngineer from specifications gathered from the company. Each drawing is printed on an 8.5 in 11 in sheet of paper. Drawings of both the solid-model type and the wireframe type were included in the stack to show the assembly in a composite mode and in an exploded mode. Drawings of the assembly from different perspectives were provided to present all features of the assembly. Fig. 1 shows a solid model of the air-cylinder assembly in an exploded mode. Fig. 2 shows a solid model of the air-cylinder assembly in a composite mode from two different perspectives. From one perspective (view A), the positions of two hose connectors were clearly shown, but the feature of the main shaft going through the bottom plate was not revealed. From another perspective (view B), the feature of the main shaft going through the bottom plate was revealed, but one hose connector was not shown. The drawing of an assembly in an exploded mode with all parts labeled alphabetically was placed on the top of the stack (see Fig. 1). Part names and part labels were printed

548

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 29, NO. 4, NOVEMBER 1999

Fig. 1. Solid-model drawing of an air-cylinder assembly in an exploded mode for the traditional engineering environment.

on this drawing. We refer to this drawing as the top assembly drawing. Drawings of the wireframe type were used to present parts, one drawing for each part. On a part drawing, the part name and the part label were also printed. The stack of drawings were arrranged in the following order: a solidmodel drawing of the assembly in an exploded mode on the top of the stack, a solid-model drawing of the assembly in an exploded mode from another perspective, two solidmodel drawings of the assembly in a composite mode for two different perspectives, two wireframe drawings of the assembly in a composite mode for two different perspectives, and the wireframe drawings of parts in the alphabetical order of part labels. Provided with a stack of drawings, subjects were allowed to re-sort the drawings and view drawings from different angles. For the DVR and CVR conditions, the solid model of the assembly was displayed in a composite mode and in an exploded mode (see Figs. 3 and 4). A hard copy of the solid model drawing of the assembly in an exploded mode

(the top drawing of the assembly used in the TE condition) also was provided to subjects for showing part names and part labels. Subjects were allowed to rotate and move the assembly in the composite mode and the parts of the assembly in the exploded mode along certain axes. Hence, subjects were able to simulate an assembly operation by moving and putting two parts together. As two parts came close, the DVR environment or the CVR environment attached them and displayed the subassembly if two parts fit in the move direction to form their relationship as designed. If two parts cannot fit in the move direction, it goes back out to its exploded position. The DVR and CVR conditions differed in that the assembly and its parts were displayed on a 21 in color monitor in the DVR condition but were displayed in a CAVE room for the CVR condition. In the CVR environment, subjects wore stereo shutter glasses and used a wand instead of a mouse to make use of 3-D immersion in the CAVE room. In addition, the DVR environment did not provide user-centered perspective,

YE et al.: A COMPARATIVE STUDY OF ASSEMBLY PLANNING

549

Fig. 3. Presentation of an air-cylinder assembly in a composite mode and in an exploded mode for a nonimmersive, desktop VR environment. Fig. 2. Solid-model drawing of an air-cylinder assembly in a composite mode from two different perspectives for the traditional engineering environment.

C. Subjects

whereas the CVR condition did. Using the provision in the CVR environment, subjects were able to view the assembly inside–out if necessary. For example, subjects could dive into the inside of a part of the assembly to view the exact fit of the assembly, which was not possible with the DVR environment. In summary, the TE condition showed all features of the assembly and its parts using a limited number of perspectives, each perspective on a separate sheet. The DVR and CVR conditions presented all features of the assembly and its parts based on specified perspectives by subjects, all in one place. The perspectives in both the TE condition and the DVR condition presented the assembly from the outside. The CVR condition provided user-centered perspectives of the assembly and its parts from both the outside and the inside, all in one place.

Fifteen subjects from the University of Illinois, Chicago participated in the experiment. They were graduate students and college students (juniors and seniors) with backgrounds in industrial engineering, mechanical engineering, electrical engineering, and art. Among the subjects were four females and eleven males. The subjects participated in the experiment on a voluntary basis. Except for one subject, all of the subjects stated that they had no experience in assembly planning prior to the experiment. There were five subjects for each experimental condition (TE, DVR, or CVR). The subjects in the TE condition had an average age of 24.8 with a standard deviation of 2.68. The subjects in the DVR condition had an average age of 26.8 with a standard deviation of 3.96. The average age of the subjects in the CVR condition was 25.0 with a standard deviation of 4.00.

550

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 29, NO. 4, NOVEMBER 1999

Fig. 4. Presentation of an air-cylinder assembly in a composite mode and in an exploded mode for an immersive CAVE VR environment.

D. Experimental Task The subjects were asked to generate an assembly sequence for the air-cylinder assembly that was presented in their assigned environment. For the TE condition, the subjects were asked to review the stack of drawings for the assembly and its parts, generate an assembly sequence for the assembly, and specify the assembly sequence on the provided sheet. The specification of an assembly sequence included the following items of information for each assembly operation: the step number, the description of the assembly operation, and the part moved during the assembly operation. The subjects were allowed to view the drawings from different angles and re-sort the drawings. For the DVR condition, the subjects were asked to review the assembly and its parts on the computer screen, assemble the parts in the exploded mode of the assembly by rotating and moving the parts while generating an assembly sequence for the assembly, and specify the assembly sequence on the provided sheet. The subjects were allowed to rotate the assembly in the composite mode and parts in the exploded mode of the assembly for viewing them from different perspectives.

In the CVR condition, the subjects were asked to perform the assembly planning task in a manner similar to that in the DVR condition. The major difference between the DVR condition and the CVR condition lies in the way in which the assembly and its parts were viewed (nonimmersive versus immersive). For each of the three conditions, the subjects were instructed that there was no time limit on the assembly planning task. However, the subjects were asked to complete the assembly planning task in the shortest time possible without sacrificing the quality of their assembly sequence. They were told that the quality of their assembly sequence would be evaluated based on the feasibility and goodness criteria. E. Experimental Procedure The subjects were asked to complete the experiment in the following steps. 1) Read and sign an informed consent form. 2) Fill out a pre-experiment questionnaire. 3) Complete the training and practice on assembly planning.

YE et al.: A COMPARATIVE STUDY OF ASSEMBLY PLANNING

4) Perform an assembly planning task in the experimental session. 5) Fill out a postexperiment questionnaire. From the pre-experiment questionnaire, information such as the subjects’ age, gender, and experience with assembly planning was obtained. Before the experimental session, the subjects received training on assembly planning and their assigned environment. The subjects were asked to read a two-page document, which introduced them to assembly planning and illustrated assembly planning using an example assembly. The assembly was a reduction gear assembly consisting of seven parts. The material showed a good assembly sequence for the assembly. In the material, the feasibility and goodness criteria of evaluating an assembly sequence were explained. A feasible assembly sequence prevents part collisions and part trappings that are related to the geometric features and topological structure of an assembly and its parts [6]–[12]. In addition to preventing part collisions and part trappings, a good assembly sequence minimizes the reorientation and difficulty of handling parts and maximizes the efficiency and safety of handling parts [6]–[12]. Four specific goodness criteria were considered in this study: reorientation, handling difficulty, similarity, and stability. The reorientation criterion measures the number of excessive parts or assembly reorientation. The difficulty criterion measures the number of difficult assembly operations. The similarity criterion measures whether similar parts (e.g., four rods in the air-cylinder assembly) are put together in a similar manner for efficiency. The stability criterion measures how tightly assembled parts stay together without falling apart. These goodness criteria are aimed at improving the efficiency of assembly operations and preventing damages to parts while performing assembly operations. The feasibility and goodness criteria were demonstrated to the subjects through a bad assembly sequence for the reduction gear assembly. Feasibility and goodness problems of this assembly sequence were explained in the material. The subjects were encouraged to ask questions about the material. After reading the material, the subjects were asked to practice an assembly planning task in their assigned condition for a gear assembly consisting of eleven parts. The purpose of this practice was to let the subjects get familiar with: 1) the methods to view and manipulate an assembly in their assigned condition; 2) the generation of an assembly sequence based on the considerations of feasibility and goodness criteria, and 3) the specification of an assembly sequence. After a subject generated an assembly sequence, the experimenter examined it and pointed out to the subject any problems with respect to the feasibility and goodness criteria. The subject was then asked to revise the assembly sequence. This process continued until the subject generated a good assembly sequence. Then the subject was allowed to proceed to the experimental session. Therefore, through this practice, the subject’s understanding of assembly planning was tested and insured. In the training session, the subjects were allowed to ask questions at any time. In the experimental session, the subjects performed an assembly planning task for the air-cylinder assembly (see

551

Fig. 5. Good assembly sequence for an air-cylinder assembly.

Fig. 1) in a similar manner as in the practice during the training session. However, in the experimental session, the subjects must work independently without the question–answer interactions with the experimenter. In the experimental session, no feedback was provided by the experimenter to the subjects regarding the problems with their assembly sequence. The assembly sequences generated by the subjects in the experimental session were later analyzed to collect the subjects’ performance data. After completing the assembly planning task in the experimental session, the subjects were asked to fill out the postexperiment questionnaire. The subjects were provided with rating scales from one to seven to rate their understanding of the air-cylinder assembly (one for “well” and seven for “poor”), difficulty in determining the feasibility aspect of their assembly sequence (one for “easy” and seven for “difficult”), and difficulty in determining the goodness aspect of their assembly sequence (one for “easy” and seven for “difficult”). III. RESULTS A. Subject Performance Two categories of performance data were collected from each subject: performance time and performance quality. The time that each subject took to complete the assembly planning task in the experimental session was recorded. To collect performance quality data, each subject’s assembly sequence was analyzed to obtain seven measures: NIF The number of infeasible assembly operations. NER The number of assembly operations requiring excessive reorientation. ND The number of difficult assembly operations. NUS The number of unstable assembly operations. NDS The number of dissimilar assembly operations for similar parts. NM The number of missing parts. TN The total number of problematic assembly operations which was the sum of the other measures. These seven measures were obtained by comparing a subject’s assembly sequence with a good assembly sequence for the air-cylinder assembly that we generated (see Fig. 5). Each step in a subject’s assembly sequence was marked OK, infeasible, excessively reoriented, dissimilar, or unstable by comparing the assembly operation of installing a part in that

552

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 29, NO. 4, NOVEMBER 1999

TABLE I DESCRIPTIVE STATISTICS OF EXPERIMENTAL RESULTS

step with the assembly operation of installing the same part in the good assembly sequence. For example, if a subject’s assembly sequence contained the sequential steps Q, G, and M, the step of installing M would be marked infeasible. An excessive reorientation was required to install D’s in the sequential steps of Q, E, D, D, F, D, and D as compared to the way in which D’s were installed in the good assembly sequence. In the sequential steps of V, U, and T, the step of installing T would be marked difficult. In the sequential steps of Q, R, S, T, U, L, N, M, and G, the step of installing L would be marked dissimilar in the way of installing L, which is similar to U on the other side of Q. In the sequential steps of Q, E, and F, the step of installing E would be marked unstable because it formed a loose contact with Q and might fall apart when we work on F. If part B was not specified in an assembly sequence, a step was considered missing. It should be noted that there exist a number of assembly sequences equivalent to the good assembly sequence shown in Fig. 5 with respect to the quality of assembly sequences. The air-cylinder assembly has the symmetrical structure centered around part Q (the center plate). The same group of parts was added to both sides of the center plate. The assembly sequence could start from either side of the center plate, which yielded the same level of feasibility and goodness measures for either assembly sequence. For example, the sequential steps of Q, O, N, M, G, R, S, and T were equivalent to the sequential steps of Q, O, R, S, T, N, M, and G. After marking each step of a subject’s assembly sequence, the number of infeasible steps, excessive reorientation steps, difficult steps, dissimilar steps, unstable steps, and missing steps were counted. Values were calculated for variables NIF, NER, ND, NUS, NDS, NM, and TN. Table I shows the descriptive statistics of these variables and performance time by the three conditions. Using the SAS’s generalized linear models (GLM) procedure [39], an analysis of variance (ANOVA) was performed to analyze the effects of the three conditions on each of these variables. The effect of the three conditions on the subjects’ per. formance time was significant A Duncan’s multiple comparison test [39] revealed that the performance time of the TE condition was significantly longer than the DVR and CVR conditions, whereas the difference in the performance time between the DVR condition and the CVR condition was not significant. There were significant effects of the three conditions on , ND NER , NDS , and TN

. A Duncan’s test on NER revealed that the subjects’ assembly sequences in the TE condition involved significantly more reorientations than in the DVR condition, and that differences for other comparisons (TE versus CVR, CVR versus DVR) were not significant. A Duncan’s test on ND revealed that the number of difficult steps in the TE condition was significantly greater than the number of difficult steps in the DVR condition, and that the number of difficult steps in the DVR condition also was significantly greater than the number of difficult steps in the CVR condition. A Duncan’s test on NDS revealed that the number of dissimilar steps in the TE condition was significantly greater than the number of dissimilar steps in the CVR condition, and that the number of dissimilar steps in the CVR condition was significantly greater than the number of dissimilar steps in the DVR condition. A Duncan’s test on TN revealed that the total number of problematic steps in the TE condition was significantly greater than the number of problematic steps in the CVR condition, and that differences for other comparisons (TE versus DVR, DVR versus DVR) were not significant. The effects of the three conditions on NIF, NUS, and NM were not significantly different. They were for NIF, for NUS, and for NM. B. Subjective Ratings Three subjective rating values were collected for each subject directly from the post-task questionnaire. RU Rating for assembly understanding. RF Rating for meeting the feasibility criterion. RG Rating for meeting the goodness criteria. TR Total rating, which was the sum of RU, RF, and RG. A lower rating value indicated a better understanding of the assembly or less difficulty in meeting the feasibility or goodness criteria. Table I shows the descriptive statistics of the subjective ratings. ANOVA’s for the effects of the three conditions on RU, RF, RG, and TR revealed no significant for RU, effects for RF, for RG, and for TR). Based on the subjective evaluation of the three conditions, the subjects did not significantly favor one condition over another. IV. DISCUSSIONS Based on the above results, it became apparent that the subjects in the TE condition performed as well as the subjects

YE et al.: A COMPARATIVE STUDY OF ASSEMBLY PLANNING

in the DVR and CVR conditions in meeting the feasibility criterion. The difference between the TE environment and the two VR environments was found mainly in meeting goodness criteria, which contributed to the overall performance differences as reflected in the effects of the three conditions on performance time and TN. Among the various goodness criteria, criteria of reorientation, handling difficulty, and similarity benefited significantly from using the two VR environments for assembly planning. The CVR condition outperformed the DVR condition on the criterion of handling difficulty, whereas the DVR condition outperformed the CVR condition on the similarity criterion. There was no significant difference between the CVR condition and the DVR condition on the criterion of excessive reorientation. Overall, the performance difference between the DVR environment and the CVR environment was not significant, as reflected in the performance time and the total number of problematic steps TN. An examination of the subjects’ assembly sequences from the TE condition revealed some reasons underlying the subjects’ inferior performance in the TE condition. A common problem among the subjects in the TE condition was that all the subjects used a stack-up method to generate their assembly sequence from the bottom plate (part V) to the top plate (part K). This stack-up order seemed relevant to the way in which the parts were presented in the top assembly drawing (see Fig. 1). Since this drawing included the names and labels of all the parts, it is likely that it was used by the subjects as the main reference document while generating their assembly sequence. The parts in this drawing appeared stacked up on the bottom plate (part V). However, the bottom plate should not be used as the supporting base of the assembly operations, because one end of the main shaft (part O) went through a hole in the bottom plate, making the assembly unstable when standing on the bottom plate. This design detail was illustrated in other drawings of the assembly in the composite mode. Although the subjects were provided with different drawings for different perspectives of the assembly and were allowed to view each drawing from any angle if necessary, their assembly planning appeared to be dominated by the perspective and angle from which the assembly was illustrated in the top assembly drawing. The subjects failed to recognize the symmetric structure of the assembly. Moreover, some subjects failed to recognize the position of the hole on the connector holder (part E) where the plug (part B) was inserted. This position was not obvious in the top assembly drawing, but became apparent from the assembly drawings in a composite mode and the part drawing of the connector holder. The stack-up method caused a number of problems with the subjects’ assembly sequences. First of all, the installation of the aluminum cylinder (part U) onto the bottom plate before other parts (e.g., inner locknut, pressure plate, and pressure seal) made it difficult to install those other parts that went inside the aluminum cylinder. Secondly, the installation of inner lockout, pressure plate, pressure seal, and center plate (parts T, S, R, Q) in a stacked-up order made it infeasible to lock the main shaft onto the inner locknut. Because the main shaft went through the bottom plate, a reorientation of the

553

already stacked-up subassembly (including the bottom plate, aluminum cylinder, inner locknut, pressure plate, pressure seal, and center plate) was required when installing the main shaft through the already stacked-up subassembly. Finally, the stack-up method also caused the application of different assembly operations to the similar groups of parts on both sides of the center plate. The stack-up problem in the TE condition might be attributed to the lack of a coherent view of the assembly. In the TE condition, different perspectives of the assembly were presented in different drawings on separate sheets. Hence, in order to obtain a comprehensive understanding of the assembly, subjects might have to construct a mental model of the assembly, including all features from disjoint, partial views of the assembly. It also might be difficult for the subjects to check the correctness of their mental model of the assembly. In contrast, in the DVR and CVR conditions, the physical model of the assembly was presented for the subjects to explore features all in one place. The smooth transitions among different perspectives in the two VR conditions might provide the subjects with an advantage in understanding the assembly. The manipulation of the physical model of the assembly also might give the subjects in the two VR environments an easy opportunity to check the correctness of their assembly understanding. These advantages of the VR environments might help the subjects in discovering the symmetrical structure and other features of the assembly, which might in turn improve their assembly planning. Even if there were errors in the subjects’ understanding of the assembly, the simulation of the assembly sequence in the VR environments might help the subjects in recognizing those errors when difficulty was encountered during the simulation. However, in the TE condition, the subjects had to carry out the understanding of the assembly and the verification of their assembly sequence mentally without the assistance of the physical model of the assembly and the physical simulation of the assembly sequence. Therefore, the physical model of the assembly and the simulation of the assembly sequence in the VR environments might help in reducing the mental workload of the subjects in assembly planning, thus leading to less errors in assembly planning. Based on the insignificant effects of the three conditions of NIF, NUS, and NM, we found that the two VR environments did not produce improvements over the TE environment in reducing the number of infeasible assembly operations, the number of unstable assembly operations, and the number of missing parts. With respect to the insignificant effect of NUS, an examination of the subjects’ assembly sequences from the three conditions revealed that the subjects attempted to manipulate multiple parts at the same time instead of installing parts one by one. For example, pressure plate and pressure seal (parts S and R or parts M and N) were often put together into a subassembly by the subjects, and then the subassembly (in a loose state) was placed on the center plate. The subassembly in a loose state was unstable, which created difficulty in performing the assembly operation of placing them on the center plate. Although the CVR environment provided the subjects with a more realistic visualization of the assembly

554

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 29, NO. 4, NOVEMBER 1999

and its parts, the CVR environment still differed from the real world in many other aspects. For example, gravity exists in the real world, whereas in the CVR environment of this study, the subjects could not feel the effect of gravity and part weight. A provision of gravity and part weight in the CVR environment might help the subjects in meeting the stability criterion. The subjects’ performance in meeting the feasibility criterion might be improved by incorporating a rigid collision detection into the VR environments to reveal part collisions and part trappings during the simulation of an assembly sequence. Mechanisms (e.g., an indexing system) must be developed to prevent the subjects from missing parts in an assembly sequence. In conclusion, the results of this study revealed certain advantages of the VR environments over the traditional engineering environment in supporting assembly planning. Especially a coherent, flexible visualization and manipulation of the assembly and its parts in one place on the computer produced improvements of the subjects’ overall performance and in meeting some goodness criteria such as handling difficulty, excessive reorientation, and similarity. Although the enhanced visualization and simulation are necessary to support assembly planning, other factors such as gravity, part weight, flexibleshape parts, and irregular geometry also may be helpful in bringing the subjects’ assembly planning performance in the VR environments to a higher level. The immersive CVR environment is a more natural place to incorporate these factors than the nonimmersive DVR environment. When these factors are incorporated into the CVR environment, we may be able to observe the more dramatic advantages of the CVR environment in supporting assembly planning. In this study, the differences between the CVR environment and the DVR environment were not significant with respect to the performance time and the total number of problematic steps, despite the fact that one was better than the other for one goodness measure but was worse for another goodness measure. Considering the advantages of the VR environments in addressing the goodness aspect of assembly planning, the VR environments could complement the more algorithmic approaches by presenting feasible assembly steps from those approaches to engineers for evaluation and selection based on goodness criteria. It should be noted that the results of this study were obtained from an experiment using a specific assembly. The nature of a controlled experiment prevents us from exhausting all kinds of assembly in a controlled experiment. This may place some limitation on generalizing the results of this study. For example, the assembly for this study had a stacking structure, which in turn determined the assembly sequence in a linear manner. This assembly planning task did not involve some other issues relating to fragile parts, flexible parts, irregular geometry, fixtures, etc. Further study can be carried out to classify features of assembly structure and operation, and to conduct experiments using two independent variables: one for different assembly features and another for different environments. From this study, we cannot draw conclusions on how VR advantages are related to assembly

features. Nevertheless, the assembly for this study is a real product. It has a lot in common with many real-world assembly products in assembly structure and operation. Hence, the results of this study on this specific assembly should promote the consideration of using VR environments to support the goodness aspect of assembly planning. It also should be noted that almost all of the subjects for this study had no experience in assembly planning prior to the experiment in order to provide an unbiased comparison of the environments studied. However, in practice, engineers often work on similar assembly products, and thus possess considerable knowledge of assembly products and their assembly plan. In such cases, the advantages of the VR environments with respect to the TE environment may be reduced considerably due to the engineers’ familiarity with assembly products. Further study is needed to investigate how VR environments could complement the knowledge and expertise of engineers in reducing cognitive overload of engineers to produce feasible and good assembly plans efficiently. ACKNOWLEDGMENT The authors would like to thank N. Ahmed for running a part of this experiment and S. Anantharaman for developing the assembly and part drawings in ProEngineer. We also would like to thank the anonymous reviewers for their constructive and helpful comments. REFERENCES [1] M. P. Groover and E. W. Zimmers, CAD/CAM: Computer-Aided Design and Manufacturing. Englewood Cliffs, NJ: Prentice-Hall, 1984. [2] U. Rembold, B. O. Nnaji, and A. Storr, Computer Integrated Manufacturing and Engineering. Wokingham, U.K.: Addison-Wesley, 1993. [3] G. Salvendy, Handbook of Industrial Engineering. New York: Wiley, 1982. [4] N. Singh, Systems Approach to Computer-Integrated Design and Manufacturing. New York: Wiley, 1996. [5] I. Zeid, CAD/CAM Theory and Practice. New York: McGraw-Hill, 1991. [6] S. Chakrabarty and J. Wolter, “A structure-oriented approach to assembly sequence planning,” IEEE Trans. Robot. Automat., vol. 13, pp. 14–29, Feb. 1997. [7] A. Delchambre, Computer-Aided Assembly Planning. London, U.K.: Chapman & Hall, 1992. [8] L. S. H. de Mello and A. C. Sanderson, “Two criteria for the selection of assembly plans: Maximizing the flexibility of sequencing the assembly tasks and minimizing the assembly time through parallel execution of assembly tasks,” IEEE Trans. Robot. Automat., vol. 7, pp. 626–633, Oct. 1991. [9] L. Laperriere and H. A. ElMaraghy, “GAPP: A generative assembly process planner,” J. Manuf. Syst., vol. 15, no. 4, pp. 282–293, 1996. [10] K. T. Seow and R. Devanathan, “A temporal framework for assembly sequence representation and analysis,” IEEE Trans. Robot. Automat., vol. 10, pp. 220–229, Apr. 1994. [11] N. Ye and D. A. Urzi, “Heuristic rules and strategies of assembly planning: Experiment and implications in the design of assembly decision support system,” Int. J. Prod. Res., vol. 34, no. 8, pp. 2211–2228, 1996. [12] U. A. Seidel, “MI system for assembly sequence planning: Human factors consideration,” in Design for Manufacturability, M. Helander and M. Magamachi, Eds. London, U.K.: Taylor & Francis, 1992. [13] T. Cao and A. C. Sanderson, “Task decomposition and analysis of robotic assembly task plans using Petri nets,” IEEE Trans. Ind. Electron., vol. 41, no. 6 pp. 620–630, Dec. 1994. [14] C. L. P. Chen, “Automatic assembly sequences generation by patternmatching,” IEEE Trans. Syst., Man, Cybern., vol. 21, pp. 376–389, Apr. 1991. [15] C. L. Chen and C. A. Wichman, “A systematic approach for design and planning of mechanical assemblies,” AI EDAM, vol. 7, no. 1, pp. 19–36, 1993.

YE et al.: A COMPARATIVE STUDY OF ASSEMBLY PLANNING

[16] L. S. H. de Mello and A. Sanderson, “A correct and complete algorithm for the generation of mechanical assembly sequences,” IEEE Trans. Robot. Automat., vol. 7, pp. 228–240, Feb. 1991. [17] Y. F. Huang and C. S. G. Lee, “A framework of knowledge-based assembly planning,” in Proc. IEEE Int. Conf. Robotics and Automation, 1991, pp. 599–604. [18] J. L. Hwang and M. Henderson, “Applying the perception to 3-D feature recognition,” in Proc. 1992 NSF Design and Manufacturing Grantees Conf., pp. 703–707. [19] E. Knoll, “Spatial reasoning in assembly planning,” in Proc. 1992 NSF Design and Manufacturing Grantees Conf., 1089–1097. [20] E. Knoll and R. Mohammad, “Modeling and reasoning for assembly planning,” in Proc. 1992 NSF Design and Manufacturing Grantees Conf., pp. 721–725. , “Spatial reasoning about three-dimensional mechanical assem[21] blies,” in Proc. 1994 NSF Design and Manufacturing Grantees Conf., pp. 73–74. [22] R. Mohammad and E. Knoll, “Automatic generation of exploded views by graph transformation,” in Proc. 1993 NSF Design and Manufacturing Grantees Conf., pp. 1425–1426. [23] S. Lee, “Backward assembly planning,” in Artificial Intelligence Applications in Manufacturing, A. Famili, D. S. Nau, and S. H. Kim, Eds. Menlo Park, CA: AAAI Press, 1992, pp. 61–102. [24] A. C. Lin and T.-C. Chang, “3-D MAPS: Three-dimensional mechanical assembly planning system,” J. Manuf. Syst., vol. 12, no. 6, pp. 437–456, 1993. [25] H. C. Liu and B. O. Nnaji, “PAM: A product modeler for assembly,” in Proc. 1994 NSF Design and Manufacturing Grantees Conf., pp. 69–70. [26] Y. Liu and R. Popplestone, “Symmetry groups in solid model-based assembly planning,” in Artificial Intelligence Applications in Manufacturing, A. Famili, D. S. Nau, and S. H. Kim, Eds. Menlo Park, CA: AAAI Press, 1992, pp. 61–102. [27] J. Wolter, “Automatic generation of assembly plans,” in Proc. IEEE Int. Conf. Robotics and Automation, 1989, pp. 62–68. [28] J. Wolter, S. Chakrabarty, and J. Tsao, “Methods of knowledge representation for assembly planning,” in Proc. 1992 NSF Design and Manufacturing Grantees Conf., pp. 463–468. [29] R. D. Caldwell, N. Ye, and D. A. Urzi, “Re-engineering the product development cycle and future enhancements of computer-integrated manufacturing environment,” Int. J. Comput. Integrated Manuf., vol. 8, no. 6, pp. 441–447, 1995. [30] D. F. Baldwin, T. E. Abell, M.-C. M. Lui, T. L. DeFazio, and D. E. Whitney, “An integrated computer aid for generating and evaluating assembly sequences for mechanical products,” IEEE Trans. Robot. Automat., vol. 7, pp. 78–94, Feb. 1991. [31] S. Leong and C. McLean, “Application of simulation tools in manufacturing engineering,” NIST Document, Manufact. Syst. Integrated Div., Gaithersburg, MD, 1996. [32] S. Ressler, “Applying virtual environment to manufacturing,” NIST Document 5343, Manufact. Syst. Integrated Div., Gaithersburg, MD, 1994. [33] R. D. Wilson, “Minimizing user queries in interactive assembly planning,” IEEE Trans. Robot. Automat., vol. 11, pp. 308–312, Apr. 1995. [34] U. Roy, P. Banerjee, and C. R. Liu, “Design of an automated assembly environment,” Comput.-Aided Des., vol. 21, no. 9, pp. 561–569. [35] W. Barfield and R. Robless, “The effects of two- or three-dimensional graphics on the problem-solving performance of experienced and novice decision makers,” Behavior Inform. Technol., vol. 8, pp. 367–385, 1989. [36] J. E. Fowler and M. E. Luce, “Systems integration for manufacturing applications program 1995 annual report,” NIST Document 5839, Manufact. Syst. Integrated Div., Gaithersburg, MD, 1996. [37] J. Hartman and J. Warnecke, 1996. The VRML 2.0 Handbook. Reading, MA: Addison-Wesley, 1996. [38] D. Pape, C. Cruz-Neira, and M. Czernuszenko, 1997, “CAVE User’s Guide,” Elect. Visualization Lab., Univ. Illinois, Chicago. [39] SAS Institute Inc., SAS/STAT User’s Guide, Version 6, 4th ed. Cary, NC: SAS Institute, 1990.

555

Nong Ye was born in Beijing, China, in 1964. She received the B.S. degree in computer science from Peking University, Beijing, the M.S. degree in computer science from the Chinese Academy of Sciences, Beijing, and the Ph.D. degree in industrial engineering from Purdue University, West Lafayette, IN. Since 1998, she has been an Associate Professor with the Department of Industrial Engineering, Arizona State University, Tempe. From 1994 to 1998, she was an Assistant Professor with the University of Illinois, Chicago. From 1991 to 1994, she was an Assistant Professor with Wright State University, Dayton, OH. Her research interests are in information technology, system engineering and its support to engineering design and integration, distributed virtual enterprises, and information security. Dr. Ye is a senior member of the Institute of Industrial Engineers. She serves on the editorial boards of the International Journal of Human–Computer Interaction and the International Journal of Cognitive Ergonomics.

Prashant Banerjee received the B.Tech degree from the Indian Institute of Technology, Kanpur, India and the M.S. and Ph.D. degrees in industrial engineering from Purdue University, West Lafayette, IN. He is currently an Associate Professor in the Department of Mechanical Engineering, University of Illinois, Chicago, while also serving as the Director of the Industrial Virtual Reality Institute (IVRI), a joint research and development operation comprised of the University of Illinois, Chicago, Northwestern University, Chicago, IL, and Argonne National Laboratory, Argonne, IL. His current research interests include virtual reality-based factory design, part design and assembly design models, immersive display interfaces, and linear and nonlinear design optimization models. His research has been supported by NSF, NIST, and ONR and by companies such as Caterpillar, Searle, and Motorola,. Dr. Banerjee is currently serving as Department Editor of IEE Transactions and as an Associate Editor of IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION.

Amarnath Banerjee received the M.Sc.(Tech) degree in computer science from the Birla Institute of Technology and Science, Pilani, India. He currently is pursuing the Ph.D. degree in Industrial Engineering in the Department of Mechanical Engineering, University of Illinois, Chicago. His research interests include telecollaboration in virtual environments, simulation of object behavior in virtual reality, and task integration.

Fred Dech, photograph and biography not available at the time of publication.