James L. Mathieson

Complexity Metrics for Directional Node-Link System Representations: Theory and Applications

This paper presents an approach to defining and quantifying the complexity of systems as represented in mixed (directed and non-directed) bipartite graphs through the presentation of a central example as well as other applications. The approach presented defines nine measurements of different properties of the graph system. These measurements are derived from the representation of the system into a three dimension relational design structure matrix as well as the projections and transformations of this matrix. The metrics generated address dimensional and connective size, shortest path properties, and the decomposability of the system. Finally, a normalization and aggregate approach of these metrics is then given. This aggregation is visualized with spider graphs that facilitate viewing multiple aspects of complexity within a single perspective.Copyright

Assembly time modelling through connective complexity metrics

This paper presents the development of a model for predicting the assembly time of a system based on complexity metrics of the system architecture. A convention for modeling architecture is presented, followed by ten analyzed systems. These systems are subjected to complexity metrics developed for other applications. A model is developed based on a recognizable trend and a regression of that trend. The regression is then further refined based on its similarities to additional metrics other than that used in regression. The final model uses average path length, part count, and path length density to predict assembly time to within ±16% of that predicted by the Boothroyd and Dew Hurst design for assembly analysis method.

COMPLEXITY AS A SURROGATE MAPPING BETWEEN FUNCTION MODELS AND MARKET VALUE

The purpose of this paper is to investigate if early stage function models of design can be used to predict the marketvalue of a commercial product. In previous research, several metrics of complexity of graph-based product models have been proposed and suitably chosen combinations of these metrics have been shown to predict the time required in assembling commercial products. By extension, this research investigates if this approach, using new sets of combinations of complexity metrics, can predict market-value. To this end, the complexity values of function structures for eighteen products from the Design Repository are determined from their function structure graphs, while their market values are procured from different vendor quotes in the open market. The complexity and value information for fourteen samples are used to train a neural net program to define a predictive mapping scheme. This program is then used to predict the value of the final four products. The results of this approach demonstrate that complexity metrics can be used as inputs to neural networks to establish an accurate mapping from function structure design representations to market values to within the distribution of values for products of similar type.

Assembly Time Modeling through Connective Complexity Metrics

This paper presents the development of a model for predicting the assembly time of a system based on complexity metrics of the system architecture. A convention for modeling architecture is presented, followed by ten analyzed systems. These systems are subjected to complexity metrics developed for other applications. A model is developed based on a recognizable trend and a regression of that trend. The regression is then further refined based on its similarities to additional metrics other than that used in regression. The final model uses average path length, part count, and path length density to predict assembly time to within ±16% of that predicted by the Boothroyd and Dew Hurst design for assembly analysis method.

Representation: Structural Complexity of Assemblies to Create Neural Network Based Assembly Time Estimation Models

Assembly time estimation is traditionally a time intensive manual process requiring detailed geometric and process information to be available to a human designer. As a result of these factors, assembly time estimation is rarely applied during early design iterations. This paper explores the possibility that the assembly time estimation process can be automated while reducing the level of design detail required. The approach presented here trains artificial neural networks (ANNs) to estimate the assembly times of vehicle sub-assemblies at various stages using properties of the connectivity graph at that point as input data. Effectiveness of estimation is evaluated based on the distribution of estimates provided by a population of ANNs trained on the same input data using varying initial conditions. Results suggest that the method presented here can complete the time estimation of an assembly process with +/− 15% error given an initial sample of manually estimated times for the given sub-assembly.Copyright

Information Generation in the Design Process

This paper proposes and demonstrates a protocol for measuring information generated throughout a design process. The intent is to provide a consistent approach to allow the comparison of different design procedures and processes. The proposed method divides the design process into requirement, function, and component domains occurring within design iterations. To measure information or complexity in each of these domains, the elements describing the domains are counted and their mappings within and across the domains are computed. The results show that the proposed protocol and information metrics produce data points of comparable order across all domains under different design situations. Furthermore, it is shown that within-domain-coupling and across-domain-coupling metrics should be accommodate the continual increase in element count size without hiding relative changes in information generation throughout the process. When this correction is applied, it is observed that across-domain-coupling displays a decaying process of converging and diverging towards a steady state level. This presents possible support for the concepts of modeling the design process as a series of convergent and divergent processes while also suggesting that such oscillation may not be necessary.Copyright

Limitations to Function Structures: A Case Study in Morphing Airfoil Design

This paper presents an application of function structures in the design of single component morphing airfoils. A series of function structures are developed in an attempt to create an accurate model of a passively morphing airfoil. This paper describes each approach taken, while noting their modeling aspects that were successful or unsuccessful to identify representation features of existing function structure syntax and rules that relate to the usability of the models. System boundaries and definitions, function and flow definitions, carrier flows, and system state definitions are investigated. It was found that function structures in their current state, are incapable of accurately modeling the functionality of the shape changing aspect of a passively morphing airfoil. These limitations are identified and recommendations necessary to model artifacts similar to the morphing airfoil are presented. Finally, previous research efforts in the design of morphing airfoil technology are presented as solutions to certain key functions identified, thereby demonstrating the possible benefits of implementing function structures to this field.© 2010 ASME

A protocol for modeling and tracking engineering design process through structural complexity metrics applied against communication networks

A protocol for tracking the interconnection of design communication elements as a mixed temporal hypergraph network which evolves in real time is presented here. The protocol uses email and limited human reporting data to develop the time-stamped connections of the network. At any time, this network, or a filtered subset, may be subjected to an analysis of graph and network complexity properties. The response of these properties may then be correlated to either events, such as design stages, or performance metrics, such as designer effort. This approach is applied to emails generated in the course of an undergraduate mechanical engineering senior design project demonstrating an ability to identify member roles, work schedules, and project phase changes from the network properties.

Evaluation of System-Directed Multimodal Systems for Vehicle Inspection

ed the inspection task. Three concentric geometric shapes comprised one symbol that simulated a checkpoint, called an “inspection item marker” (Fig. 5(b)). The inspection item markers were used as indicators for the location and status of the item and how to perform inspection. The outermost shape, called the “location indicator,” matched a shape description on the checklist. Location indicators had three parameters: size (large, inscribed inside an 8.5 11 in. piece of paper, or small, inscribed inside a quarter-sheet of paper), color (red, yellow, green, or blue), and shape (triangle, circle, or square). For example, through text-tospeech and/or screen output, a participant is informed that the current item to check is a “Large Red Triangle.” The participant would then find that shape on the car. No two location indicators were the same on any checklist so that the location of the current checkpoint was unambiguous. Once the participant found the shape, the participant would then look at the shapes within the location indicator. The next concentric shape, called the “task indicator,” indicated how the participant should inspect the item. Depending on whether the shape is a square, pentagon, or hexagon, the participant should either look at but not touch the marker, touch the marker with only one hand, or touch the marker with two hands, respectively. Finally, within the task indicator, there was a set of four dots in a row called the “defect indicator.” The shading of these four dots indicated to the participant whether an item had a defect. An odd number or zero shaded dots indicated an item should pass inspection while an even number of shaded dots indicated a defect, which should fail inspection. In designing the abstracted task, our goal was to simulate the cognitive load and time requirements for actual vehicle inspection. We chose three-word item identifiers to roughly match the phrase length of items that occur on a typical inspection checklist and the time it would take text-to-speech to read them. Requiring the user to look at shapes to determine what action to take, and whether the item passed inspection, simulated the time and cognitive load it would take an inspector to determine the status of an actual inspection item, since in real vehicle inspection an associate must look at a part, recall how to inspect it, and then determine whether the part passes or fails inspection. Inspection item markers were placed throughout the vehicle frame to simulate the various checkpoint locations in a vehicle. The percentage of defective items per checklist approximated the defect rate reported through actual vehicle inspection at BMW. Since inspectors are provided with reference material at their stations should they forget how to inspect a particular checkpoint, we provided our participants with a lanyard holding a reference sheet for the meanings of the task and defect indicators. For each trial, there were 20 inspection item markers placed on the vehicle, providing the 15 items on the checklist plus five items as distracters, since expert inspectors would not inspect each item present on a vehicle. The participant was presented with one checklist of 15 items, ordered by their location moving counterclockwise around the vehicle. This simulated the actual inspection process, since an inspector is directed to check only certain vehicle features. Participants were given 106 s to complete the checklist, based on the time frame standard for factory inspections. If the participant finished before time was up, he or she said stop to complete the trial. If the participant did not finish in time, the system automatically stopped accepting input. We created unique checklists for ten trials. The first checklist was used for a practice trial before using any devices for input. In this first trial, the participant inspected all 20 checkpoints with the help of an experimenter to help familiarize them with the inspection item markers and the locations of the checkpoints on the vehicle. The user then completed nine trials, with three trials per device. 4.3 Measures. For each participant, we gathered data through preand postquestionnaires, a debriefing interview, Fig. 5 (a) Vehicle body used for experimental evaluation. (b) Example of shapes used for abstracted inspection task. This item would be called “Small Blue Square,” would require a one-handed touch for the inspection action since the shape is a pentagon and would pass inspection since an odd number of dots are shaded. RealVNC: http://www.realvnc.com/. Journal of Computing and Information Science in Engineering MARCH 2013, Vol. 13 / 011002-5 Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 10/04/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use experimenter logs, and software logs. The prequestionnaire gathered information about demographics, level of use of various hardware configurations, vehicle knowledge, and learning styles. We recorded the number of items inspected and not inspected, the number of items inspected correctly and incorrectly, the time it took for inspection completion, and each voice command the user spoke. While the participant conducted an inspection, an experimenter marked each checkpoint on a clipboard to indicate if the user had inspected the correct item with the correct action (look, one-hand touch, or two-hand touch). Finally, in the postquestionnaire and debriefing interview, we asked questions related to usability, preferences, and effectiveness of the interfaces and hardware configurations. 4.4 Results. Of 25 participants from Clemson University, there were 10 females and 15 males, aged 18–53 (mean1⁄4 23). Participants rated themselves as having low (N1⁄4 12), average (N1⁄4 8), and high (N1⁄4 5) levels of vehicle knowledge. Twentyfour participants were college students and one participant was a postdoctoral fellow. Since the time to complete each trial was limited to 106 s, many participants did not complete all trials for an input device. Overall, 13 participants completed all three trials using the handheld configuration (H), 14 participants completed all three trials with the large screen configuration (L), and 18 participants completed all three trials with the monocular display configuration (M). The task performance data were treated with a repeated measures 3 3 analysis of variance (ANOVA) to test for the within subject effects of hardware configurations and the within subject effects by trial. Data reported from the postquestionnaires were analyzed using the Chi-square test. The F and v tests that are reported for analysis used an alpha level of 0.05 to indicate significance. 4.4.1 Accuracy. The accuracy percentage of correctly checked items was determined by dividing the number of items that were correctly checked by the total number of items checked in each trial. Correctly checked items are those that the participant both performed the correct inspection action and reported the correct inspection result. There was no significant main effect of hardware configuration type for mean accuracy percentage of correctly checked items, F(2,38)1⁄4 1.58, p1⁄4 0.22, n 1⁄4 0:08. All configurations allowed for high accuracy with monocular display (M) as the highest and handheld (H) being the lowest. There was no significant main effect found among the sets of the trials or interaction effect of device by trial. The defect detection percentage was determined by dividing the number of defects that were correctly detected by the total number of defects for each trial. Since the number of defects varied over each trial, the total accuracy of defect detection for each trial was averaged across trials and participants, and then analyzed using a one-way ANOVA. There was no significant main effect found for the defect detection accuracy nor any significant interaction effect of device by trial. However, all configurations allowed for high accuracy as listed from highest to lowest: monocular display (M), large screen (L), and handheld (H). No significant differences were found for accuracy grouped by vehicle knowledge, device usage, or learning style. Unfortunately, no accuracy data are recorded for BMW inspectors, so we could not compare our accuracy results to the baseline accuracy achieved in the manufacturing environment. 4.4.2 Task Completion Times. Analysis revealed a significant main effect for hardware configuration type for overall task completion time. The handheld configuration (H) allowed for significantly faster overall completion time than the monocular display configuration (M) and the large screen configuration (L) (Table 1). In addition, participants’ overall performance became significantly faster by the third and last trial, F(2,38)1⁄4 4.38, p1⁄4 0.019, n 1⁄4 0:19. There was no significant interaction effect of hardware configuration by trial. A few participants discontinued the inspection task accidentally, indicating that participants were having difficulty or accidentally executed a command. A participant possibly discontinued the task accidentally if the overall task completion time was less than 105.5 s (due to rounding error) and did not check all 15 items on the list. There were eight accidental discontinuations for the handheld configuration (H) possibly due to difficulties with the touch screen interface, while there were no accidental discontinuations of the task for the monocular (M) or large screen (L) configurations, likely due to the Wizard-of-Oz setup. As a result of several participants not completing the full trial, it may be more informative to analyze completion time per individual item. This was calculated as a result of each participant’s overall time divided by the number of items each participant actually inspected. We did not find a significant main effect for hardware configuration type for task completion time per item, F(2,38)1⁄4 1.83, p1⁄4 0.18, n 1⁄4 0:09. Ho

Comparative Studies in Traction Concepts

The interaction between tires and soft soils is a complex process that has not yet been fully understood. Attempts to create analytical models which realistically simulate these interactions have proven to be exceedingly time consuming for each tire model and have achieved only limited success. Thus, the exploration and evaluation of traction concepts by analytical means is impractical. This paper posits that significantly more reliable, rapid, and cost effective development is achieved through the prototyping and experimental evaluation of traction concepts. Here, three traction prototypes are developed and evaluated by undergraduate teams in the course of an academic semester. These concepts explore the performance of grousers, inverted geometry (dimples), and cellular materials respectively using a wheel endurance and soft-soil traction testing system at Clemson University. Each concept is tested at different acceleration rates from 0–10km/h and at different loads while measuring the slip between the tire and soil surfaces. The results of experimental evaluation indicate that cellular materials present a unique slip profile which is superior to that of the two other purely geometric concepts studied. The worst performance was seen on the inverse geometry concept which presented a behavior of steadily increasing slip with respect to wheel velocity regardless of acceleration. Grousers also presented this behavior but only at higher accelerations. This suggests that not only that prototyping is preferable, but that traditional soft-soil traction approaches may be in error.Copyright