- Email: [email protected]
Design and Evaluation of a Graphical User Interface Using the Virtual Process Visualization Concept
Copyright @ IFAC Analysis. Design and Evaluation of Human-Machine Systems. Kassel. Gennany. 2001
DESIGN AND EVALUATION OF A GRAPHICAL USER INTERFACE USING THE VIRTUAL PROCESS VISUALIZATION CONCEPT Carsten Wittenberg
Siemens AG Corporate Technology User Interface Design D - 81730 Munich, Germany Tel.: +49 8963657470, Fax: +49 89 636 49428 [email protected]
Abstract: This paper presents the design and evaluation of a graphical user interface for process control of an artificial microworld. This microworld simulates a central heating system for an apartment complex. The developed user interface is based on the Virtual Process Visualization (ViProVis) method, which contains the three pillars Virtual Process Elements, Goal and State Visualization and Task-oriented Structure. In the experiments, this visualization is be compared to and evaluated against a conventional topological interface. Copyright © 20011FAC Keywords : User Interfaces, Computer Graphics, Human-Centered Design, HumanMachine Interface, Process Control, Tasks, Human Information Processing, Mental Models, Human Problem Solving
variables from among the large number of items of data is important if the current situation at any time is to be reliably and rapidly identified (Johannsen, 1993). The operator can be assisted in this situation by a graphical user interface to the technical system, suitably structured to accommodate human capabilities. The design of human-machine interfaces must be such as to assist the human's processing of information - both in terms of the storage of information and also in action planning and problem solving.
I. INTRODUCTION Complex, dynamic industrial plants are still monitored and controlled by human operators in a control room. Increasing requirements in respect of quality, economics and ecology, and increasingly efficient and inexpensive automation systems, lead to an increasing level of automation in such technical systems. As a result, the number of operators decreases, while the amount of process information that has to be monitored and controlled increases. Operators have to deal with a flood of information. Nowadays, several processes may be managed centrally from one control room, using computer-aided control systems. Because of the information preparation this requires, the operator becomes divorced from the process. Computer monitors are used to show a wealth of process information on a very limited area. However, identifying the relevant process
This paper describes the design and evaluation of a graphical user interface for process control of an artificial microworld. This user interface is based on the Virtual Process Visualization (ViProVis) method (Wittenberg, 1997, 1999, 2001) and was designed and evaluated during a research fellowship at the Laboratory of Cognitive Systems Science (University of Tsukuba, Japan). The terms of reference for this 81
symbolism, support for the model should be based on such pictorial visualization (Dutke, 1994).
research program were to extend this visualization method to new fields of applications, to generalize it.
The cognitive capacity of humans - including human operators - is limited. Complex technical systems like production processes usually contain hundreds or thousands of items of information. To ensure that the operator can control a complex technical system in every critical or non-critical situation, the structure of the process information must be function-oriented. This enables the technical process to be represented transparently.
2. REQUIREMENTS FOR PROCESS VISUAUZA TION Operators monitor and control the technical systems through the process visualization. Most of the information is presented purely visually, on various computer displays . The visualization and the displays are used as a window on the process. Depending on the quality of the information presentation, the operator is assisted to a greater or lesser degree in his daily work. An ideal process visualization can be described as a "transparent window" on the process.
3. VIRTUAL PROCESS VISUALIZATION Virtual Process Visualization is a concept which deals with the above problems by means of a special visualization technique, derived from modem computer graphics. This type of visualization can be used to show various kinds of relationships (even time-dependencies). The entire process is represented in a highly pictorial and transparent way.
On the basis of different action models of human behavior (e.g. Rasmussen, 1984; Norman, 1986), the various steps in the operators' actions can be described in simplified form : The operator observes the process information, interprets this information and deduces the state of the (sub)process. The operator compares this actual state with the optimal state - the target. If there is a relevant deviation between the actual state and the target, the operator realizes that action is required. In the last step, the operator plans and performs the necessary set of actions. User-centered process visualization should support these steps, and hence the operator.
There are three elements to this visualization concept: • virtual process elements • state and target visualization • task-oriented structures These components of the concept are described below.
3. 1 Virtual Process Elements The individual process units are implemented as socalled virtual process elements. Based on a typical member of the group of items concerned, a visual object is developed . All these graphical objects have been developed to give a consistent presentation of information and consistent interactions. In accordance with the idea of direct manipulation, operators interact directly with the process items. This visualization concept is expected to be efficient, with the process variables and the relationships between process items being visualized. Spatial variables - such as fill-up levels and flows - are shown directly, others are coded using color and shape.
It is a well known psychological fact that humans process pictorial information faster and more easily than textual information. These differences are described very well by the so-called multimodal model. This model describes different memory areas, which in each case are dependent on the modality of the initiating stimulus and possess very different characteristic abilities. The advantages of pictorial information - the so-called picture superiority effect-- are determined by this . This means that it should be advantageous if the process information to support the operator is coded pictorially.
I
0
Correct action planning by the operator is based on the mental model. The planning of action steps can be described as a dynamic cognitive simulation of the mental model. As with simulations of mathematical models on computer systems, this cognitive simulation is repeated iteratively, with the initial parameters changed each time, until a satisfactory result is achieved. The process visualization must make available the information necessary for the operator to form a correct mental model, based on system and process identification. In addition, since a mental model is usually characterized by strong
p III OFF
I
0 a)
Th ~
p "'
OK
pump symbol in
DIN 2800-1 . Pan .\
b) Vin ual process clement" PUIIII'
Figure 1: Pump-symbol Figure I shows the conventional DIN symbol for a pump and the corresponding Virtual Process Element.
82
3.2 State and Target Visualization
connected by a piping system. Using a heat exchanger (HX) the heat energy is passed on to the apartment complex, AP.
Subtasks are defined as appropriate for each higher level function, and are associated with appropriate targets. Each task is associated with a specific view, which includes 0<.0 the process items, the ,..-""~!Iii~'L_ process variables and the relationships between them . A L4 higher level overview of the rtlle·to·Al o... ~ <-.) 0 ,,,,,3 process shows O dll3 sv ' the states of the ~' system and the SII.IUlA nOt. CLOCK subtasks. The view of each subtask shows the achievement of its targets, the necessary process values, and the statuses of the process items. Because pictorial methods cannot be used to represent purely quantitative data, the process values are presented qualitatively. Threshold values and setpoints facilitate the identification of current process conditions. In the concept presented in this paper, the threshold values and setpoints are only displayed when the associated process values or process items are in abnormal states, This limits the volume of information to the essential minimum .
o repS O d,.5
L5
8''' ' ClIS,':'
Figure 2: SCARLETT (Moray et aI., 2000) The operator's task is to keep the temperature of the apartment complex as close as possible to the desired setpoint. To do this, the operator manipulates the pumps, heaters and valves. In addition to this main task, he has to carry out fault management tasks. During the operation of the system, disturbances can occur, such as leakage or breaks in piping, to which the operator must react rapidly to keep the output stable. Because of the complex relationships between the process variables, controlling this process needs a great deal of manual control work, and it is therefore a suitable application for investigating visualization concepts.
3.3 Tasks-oriented Structure To reduce the problem solving space, the process is structured hierarchically by function. This hierarchy is based on operators ' tasks, as determined in previous analyses, and the statuses of the process items concerned for the tasks. A special area on the interface shows the completion of the tasks. Each task is associated with targets. The level of achievement of the target is represented by colored bars. The color and length of the bars show whether a target has been achieved, is still within tolerance, or whether a warning or an alarm is present. From the length of a bar, the operator can determine how far the targets have been achieved, and if intervention is necessary.
Because of specific technical circumstances it was only possible to use a simplified version of SCARLETT. This new version of SCARLETT contains almost the same process elements as the original one, but malfunctions are realized in a different way. Leakage in the pipes can occur, breaks are not implemented.
4.1 The Virtual Process Elements Developed Each process contains a different set of process elements so that it is necessary to develop a special set of Virtual Process Elements. Based on previous projects (e.g. Wittenberg, 1998; Wittenberg, 1999), existing sets of process elements were extended to cover the new requirements.
4. APPLICATION
In addition to the extensions for the specific process, the functional properties were also enhanced. The new versions of Virtual Process Elements not only enable the current and desired process variables to be represented, they also indicate the direction in which the process variable should be changed. These features conform to the requirements for user-centered process-visualization (see above). These elements present the actual state, the optimal state and information about any necessary action, to assist the operator in performing his tasks.
At the Laboratory for Cognitive Systems Science (University of Tsukuba) an artificial microworld called SCARLETT (Supervisory ~ontrol ~nd Response to 1eaks: Iara at Isukuba) is used for experimental research on supervisory control (Inagaki et aI. , 1998; Moray et aI. , 2000). This microworld represents a central heating system to maintain the temperature of an apartment complex (Figure 2). It consists of three heatable reservoirs (R I, R2, R3), which are
83
green (optimal) through yellow (warning) to red (alarm). In addition, the usual symbols (e.g. a warning triangle) are displayed. The use of color and shape coding simplifies identification of the current state.
Figure 3 shows the visualization of the fill-up level in a reservoir. A small line represents the target value (50%). This line is visible only if there is a deviation from the target value. The arrows show the direction of the necessary level change. Like the "target-line" these arrows are again only displayed if there is a deviation. The size of each arrow corresponds to the deviation from the desired fill-up level.
=-iI._ Figure
This research project also encompassed the development of new visualization objects for the specific application example. For example, a visualization object was developed to show the temperature in the apartment complex (Figure 4). This object has the shape of a house (rectangle with roof). The color of this object is determined by the actual temperature. If there is a deviation from the desired temperature, a small colored line is visible. A triangle coverts this line to an arrow symbol, suggesting the necessary change oftemperature (up or down).
5: Visualization of target deviations (temperature of apartment complex) (left: actual T lower than desired T; center: actual T desired T; right: actual T higher than desired T)
Figure 3: Visualization of the actual and desired fill-up level (L) (left: actual L lower than desired L; center: actual L = desired L; right: actual L higher than desired L)
~~ D
__
Figure 6: Task-oriented interface based on the Virtual Process Visualization method (Task view: "Maintain temperature")
Figure 4: Visualization of the actual and desired temperature (T) in the apartment complex (left: actual T lower than desired T; center: actual T = desired T; right: actual T higher than desired T)
4.2 Visualization of deviations from target
qualitative
As a supplement to the objects displayed, the status of the process items is indicated by colored bars pointing in the direction of the deviation from target. If the bar is on the left-hand side (which is usually associated with a negative deviation) the process variable is too low. The state is indicted by color changes from
Figure 7: Task-oriented interface based on the Virtual Process Visualization method (Task view: "Maintain mass flow")
84
associated windows. Below the navigation buttons there are also indicators for the deviation of the actual process states from the targets. The operator can work in one task-oriented view but at the same time has an overview of all process views.
4.3 Task-oriented structure and the complete graphical user interface The main task is to control the temperature in the apartment complex to guarantee the supply of heat. Contributions to this objective are made by the reservoir heater, the heat exchanger, and pump P3.
5. EXPERIMENTS
The subordinate task is to maintain the mass flow. Without this mass flow, heat cannot be transported to the apartment complex. The mass flow is controlled by all the reservoirs, valves, pumps and pipes.
In the experiments, this visualization is be compared to and evaluated against a conventional topological interface. These experiments were intended to verifY the hypothesis that this visualization is more descriptive, thus making it easier to comprehend the actual process state.
Based on these two tasks, two process views have been defined. Figure 6 shows the view for the task "Maintain temperature in the apartment complex". This view contains mainly energybased process variables. All temperatures are color-coded, the flows through the pipes are related to the diameter of the colored pipes. The operator is given a clear view of the energy flow within the facility. The only target in this view is the desired temperature in the apartment complex.
The experiment was designed as a 2x2 matrix. Two different kinds of interface were used (Virtual Process Visualization vs. Conventional Topological Interface). The subjects (Japanese graduate students) were split into in two groups. Each group started with a different interface. After the first stage of the experiment, the type of interface was changed for the second stage. Each stage contained five different scenarios, i.e. different initial values and a different desired temperature in the apartment block.
The second task-oriented process view shows the The dependent variables were the mean errors of the task "Maintain mass flow" (Figure 7). This view temperature of the apartment block, the fill-up levels in the reservoirs, the time to respond to malfunctions contains all elements and variables which are and the number of undetected malfunctions. 30 associated with the mass flow. In this view there are further targets. The fill-up level in the different sets of variables were recorded. The variables were analyzed with the software tool reservoirs should be 50 %, leaving enough space to hold fluid flowing from preceding elements. STATISTICA. However, a fill-up level of 50 % also means that As expected, differences were found in the time to there is enough fluid stored to maintain the mass respond to malfunctions (Figure 8), the number of flow to the following process elements. Another undetected malfunctions (Figure 9) and in the mean target is associated with pump P3. This pump is error of the apartment block (Figure 10). An ANOV A located just after the heat exchanger, and it was carried out, and showed that all these results are determines the mass flow (and also the energy significant. flow) through the heat exchanger. Regardless of the current situation, there must be a flow through the heat exchanger. The flow Plot of Means (urlWelghtedl Interface Main Effec! through pump P3 must exceed a F( 1.22)=743. p<.0124 certain minimum rate. 35000,----------------------, Malfunctions like leakages are associated with the mass flow . For this reason, any loss of water is also indicated as a target deviation. Depending on the amount of the water loss, the operator can decide if there is a leakage or a break. In fact the simulation behind the interface is not able to represent breaks, but this limitation of the simulation is not known to the subject in the experiments (the operator).
-'-'
!
30000
E
25000
'0
c::
.g-
20000
.
15000
0
er 0
I-
E
;::
10000
5000
- !Conventional
Virtual
Interface
The top left area of both views Figure 8: Graph of mean values: time to respond to a malfunction includes a pane, with which the operator can navigate between the views. The navigation buttons are minimized views of the
85
There were no significant differences between the two types of interface in the mean errors of the fill-up levels_
Plot of Means (unwelghted) Interface Main Effect F(1.28)=7.00: p<.0132
07 0.6
A very interesting result was that the difference in the mean error of the temperature of the apartment complex depended on the sequence in which the interface types were used_The subjects who started with the interface based on the Virtual Process Visualization method made a significantly lower error (including when they used the conventional interface)_ Figure 11 shows a graph of these results_
Ti
0.5 0.4
i
0.3
i
0.2 0.1 00 -0.1 I
_L
-0.2 -0 3
Conventional
Vi rtual
Interface
Figure 9: Graph of mean values: undetected malfunctions 6. DISCUSSION
Plot of Means (unweighted) Interface Main Effect
F(1.28 )=6 43. p< 0171
~
3.0 , - - - - - - - - - - - - - - - - - - - - - - - - ,
"
2.8
"
~ 2.6
8 2.4
~ 2.2
-,,
t"
2.0
8.
I
1.8
I
'~"
E
1.6
,--
1.4
~ 1.2
i" t
L---------!
1.0 0.8
8 0.6 ~
04
~
0.2
i 1
I
~ 0.0' - - - - - - - - - - - - - - - - - - · ----Virtual
Conventi onal
Interface
Figure 10: Graph of mean values: mean error in the apartment complex temperature Plot of Means 2-way Interaction
F(1.28)=4 72: p<0385
u
~ 3.0 , - - - - - - - - - - - - - - -.-..
"
" 'ti.
-----,
2.8 2.6
~
2.4
~
2.2
~ 8.
2.0 1.8
:
1.6
~ 14
"~
1.2
~
0.8
~
0.4
"~
1.0
8 0.6
-i '·
t···
...............
.~ ~
_1 .•
c 0.2
~
....1... ..
c->v
0.0 ' - - - - - - - - - - - - - - - - - - - - ' Conventional
Vi rtual
ORDER
.. 0 .. ORDER ",,>c
Interface
Figure 11 : Graph of mean values: mean error in the apartment complex temperature as a function of the sequence in which the interfaces were used
86
In the experiment, two interesting points were found. The first is that there are no significant differences in the mean errors of the fill-up levels of the reservoirs. In discussion with the subjects it emerged that subjects paid the greatest attention to controlling the temperature in the apartment complex. This suggests that the hierarchy of targets was presented clearly. The other interesting fact is that there is a difference depending on the sequence in which the two interface types were used. Subjects who started with the interface based on the Virtual Process Visualization method made significantly less errors than the subjects who started with the conventional interface. This result could prove the hypothesis that a highly pictorial interface presents the process knowledge more clearly and assists in the building of a mental model (see above). Using the Virtual Process Visualization method, the technical is presented more system transparently than with the conventional technology-oriented method. The other results (mean error temperature, time to respond of to malfunctions, number undetected malfunctions) also confirm the effectiveness of this visualization concept.
Wittenberg, C (200 I): Virtuelle Prozessvisualisierung am Beispiel eines verfahrenstechnischen Prozesses. VDI-Fortschrittsbericht Reihe 22 Mensch-Maschine-Systeme Nr. 5, Diisseldorf: VDI-Verlag. ACKNOWLEDGEMENT
REFERENCES DIN 28004 Teil 3 (1988): FlieBbilder verfahrenstechnischer Anlagen; Graphische Symbole. Dutke, S. (1994): Mentale Modelle: Konstrukte des Wissens und Verstehens. Gottingen: Verlag flir angewandte Psychologie. Engelkamp, J. (1991): Das menschliche Gedlichtnis: Das Erinnern von Sprache, Bildern und Handlungen. 2. Aufiage, Gottingen: Hogrefe. Inagaki, T., N. Moray, M. Itoh (1998): Trust, Self-Confidence and Authority in HumanMachine Systems. In: Proceedings IF AC Man-Machine Systems, Kyoto, pp. 431-436. Johannsen, G. (1993): Mensch-MaschineSysteme. Berlin: Springer-Verlag. Moray, N., T. Inagaki, M. Itoh (2000): Adaptive Automation, Trust, and Self-Confidence in Fault Management of Time-Critical Tasks. Journal of Experimental Psychology, 2000, Vol. 6, No. 1, pp 44-58. Norman, D.A. (1986): Cognitive Engineering. In D.A. Norman, S.W. Draper (Eds.): User centered System Design, Hillsdale New York: Earlbaum, pp. 31-62. Rasmussen, J. (1984): Strategies for state identification and diagnosis in supervisory control tasks, and design of computer-based support systems. In: W.B. Rouse (Ed.): Advances in Man-Machine Systems Research, Volume 1, Greenwich: JAI Press, pp. 139193. Wittenberg, C. (1997): Unterstiltzung der menschlichen Informationsaufnahme durch ProzeBvisualisierung mittels virtueller ProzeBelemente. In: K.P. Glirtner (Ed.): Menschliche Zuverllissigkeit, Beanspruchung und benutzerzentrierte Automatisierung, Bonn: Deutsche Gesellschaft fur Luft- und Raumfahrt e.V. (DGLR), pp. 183-196. Wittenberg, C. (1998): Qualitative visualisation of a chemical process using the threeIn: dimensional computer graphics. Proceedings of the 17th European Annual Conference on Human Decision Making and Manual Control, France, Valenciennes: LAMIH Laboratoire d' Automatique et de Mecanique Industrielles et Humaines, Universite de Valenciennes, pp. 217-226. Wittenberg, C. (1999): Aufgabenorientierte Visualisierung eines komplexen verfahrenstechnischen Prozesses unter Verwendung dreidimensionaler Computergrafik. In: U. Arend, E. Eberleh, K. Pitschke (Eds.): Software-Ergonomie '99 Design von Informationswelten, Berichte des German Chapter of the ACM Nr. 53. Stuttgart: B.G . Teubner, pp. 335-344.
This project was made possible by the support of the Japanese Ministry of Science, Education, Sports, and Culture MONBU-SHO and the Laboratory for Cognitive System Science of Prof. T. Inagaki (University of Tsukuba, Japan).
87
DESIGN AND EVALUATION OF A GRAPHICAL USER INTERFACE USING THE VIRTUAL PROCESS VISUALIZATION CONCEPT Carsten Wittenberg
Siemens AG Corporate Technology User Interface Design D - 81730 Munich, Germany Tel.: +49 8963657470, Fax: +49 89 636 49428 [email protected]
Abstract: This paper presents the design and evaluation of a graphical user interface for process control of an artificial microworld. This microworld simulates a central heating system for an apartment complex. The developed user interface is based on the Virtual Process Visualization (ViProVis) method, which contains the three pillars Virtual Process Elements, Goal and State Visualization and Task-oriented Structure. In the experiments, this visualization is be compared to and evaluated against a conventional topological interface. Copyright © 20011FAC Keywords : User Interfaces, Computer Graphics, Human-Centered Design, HumanMachine Interface, Process Control, Tasks, Human Information Processing, Mental Models, Human Problem Solving
variables from among the large number of items of data is important if the current situation at any time is to be reliably and rapidly identified (Johannsen, 1993). The operator can be assisted in this situation by a graphical user interface to the technical system, suitably structured to accommodate human capabilities. The design of human-machine interfaces must be such as to assist the human's processing of information - both in terms of the storage of information and also in action planning and problem solving.
I. INTRODUCTION Complex, dynamic industrial plants are still monitored and controlled by human operators in a control room. Increasing requirements in respect of quality, economics and ecology, and increasingly efficient and inexpensive automation systems, lead to an increasing level of automation in such technical systems. As a result, the number of operators decreases, while the amount of process information that has to be monitored and controlled increases. Operators have to deal with a flood of information. Nowadays, several processes may be managed centrally from one control room, using computer-aided control systems. Because of the information preparation this requires, the operator becomes divorced from the process. Computer monitors are used to show a wealth of process information on a very limited area. However, identifying the relevant process
This paper describes the design and evaluation of a graphical user interface for process control of an artificial microworld. This user interface is based on the Virtual Process Visualization (ViProVis) method (Wittenberg, 1997, 1999, 2001) and was designed and evaluated during a research fellowship at the Laboratory of Cognitive Systems Science (University of Tsukuba, Japan). The terms of reference for this 81
symbolism, support for the model should be based on such pictorial visualization (Dutke, 1994).
research program were to extend this visualization method to new fields of applications, to generalize it.
The cognitive capacity of humans - including human operators - is limited. Complex technical systems like production processes usually contain hundreds or thousands of items of information. To ensure that the operator can control a complex technical system in every critical or non-critical situation, the structure of the process information must be function-oriented. This enables the technical process to be represented transparently.
2. REQUIREMENTS FOR PROCESS VISUAUZA TION Operators monitor and control the technical systems through the process visualization. Most of the information is presented purely visually, on various computer displays . The visualization and the displays are used as a window on the process. Depending on the quality of the information presentation, the operator is assisted to a greater or lesser degree in his daily work. An ideal process visualization can be described as a "transparent window" on the process.
3. VIRTUAL PROCESS VISUALIZATION Virtual Process Visualization is a concept which deals with the above problems by means of a special visualization technique, derived from modem computer graphics. This type of visualization can be used to show various kinds of relationships (even time-dependencies). The entire process is represented in a highly pictorial and transparent way.
On the basis of different action models of human behavior (e.g. Rasmussen, 1984; Norman, 1986), the various steps in the operators' actions can be described in simplified form : The operator observes the process information, interprets this information and deduces the state of the (sub)process. The operator compares this actual state with the optimal state - the target. If there is a relevant deviation between the actual state and the target, the operator realizes that action is required. In the last step, the operator plans and performs the necessary set of actions. User-centered process visualization should support these steps, and hence the operator.
There are three elements to this visualization concept: • virtual process elements • state and target visualization • task-oriented structures These components of the concept are described below.
3. 1 Virtual Process Elements The individual process units are implemented as socalled virtual process elements. Based on a typical member of the group of items concerned, a visual object is developed . All these graphical objects have been developed to give a consistent presentation of information and consistent interactions. In accordance with the idea of direct manipulation, operators interact directly with the process items. This visualization concept is expected to be efficient, with the process variables and the relationships between process items being visualized. Spatial variables - such as fill-up levels and flows - are shown directly, others are coded using color and shape.
It is a well known psychological fact that humans process pictorial information faster and more easily than textual information. These differences are described very well by the so-called multimodal model. This model describes different memory areas, which in each case are dependent on the modality of the initiating stimulus and possess very different characteristic abilities. The advantages of pictorial information - the so-called picture superiority effect-- are determined by this . This means that it should be advantageous if the process information to support the operator is coded pictorially.
I
0
Correct action planning by the operator is based on the mental model. The planning of action steps can be described as a dynamic cognitive simulation of the mental model. As with simulations of mathematical models on computer systems, this cognitive simulation is repeated iteratively, with the initial parameters changed each time, until a satisfactory result is achieved. The process visualization must make available the information necessary for the operator to form a correct mental model, based on system and process identification. In addition, since a mental model is usually characterized by strong
p III OFF
I
0 a)
Th ~
p "'
OK
pump symbol in
DIN 2800-1 . Pan .\
b) Vin ual process clement" PUIIII'
Figure 1: Pump-symbol Figure I shows the conventional DIN symbol for a pump and the corresponding Virtual Process Element.
82
3.2 State and Target Visualization
connected by a piping system. Using a heat exchanger (HX) the heat energy is passed on to the apartment complex, AP.
Subtasks are defined as appropriate for each higher level function, and are associated with appropriate targets. Each task is associated with a specific view, which includes 0<.0 the process items, the ,..-""~!Iii~'L_ process variables and the relationships between them . A L4 higher level overview of the rtlle·to·Al o... ~ <-.) 0 ,,,,,3 process shows O dll3 sv ' the states of the ~' system and the SII.IUlA nOt. CLOCK subtasks. The view of each subtask shows the achievement of its targets, the necessary process values, and the statuses of the process items. Because pictorial methods cannot be used to represent purely quantitative data, the process values are presented qualitatively. Threshold values and setpoints facilitate the identification of current process conditions. In the concept presented in this paper, the threshold values and setpoints are only displayed when the associated process values or process items are in abnormal states, This limits the volume of information to the essential minimum .
o repS O d,.5
L5
8''' ' ClIS,':'
Figure 2: SCARLETT (Moray et aI., 2000) The operator's task is to keep the temperature of the apartment complex as close as possible to the desired setpoint. To do this, the operator manipulates the pumps, heaters and valves. In addition to this main task, he has to carry out fault management tasks. During the operation of the system, disturbances can occur, such as leakage or breaks in piping, to which the operator must react rapidly to keep the output stable. Because of the complex relationships between the process variables, controlling this process needs a great deal of manual control work, and it is therefore a suitable application for investigating visualization concepts.
3.3 Tasks-oriented Structure To reduce the problem solving space, the process is structured hierarchically by function. This hierarchy is based on operators ' tasks, as determined in previous analyses, and the statuses of the process items concerned for the tasks. A special area on the interface shows the completion of the tasks. Each task is associated with targets. The level of achievement of the target is represented by colored bars. The color and length of the bars show whether a target has been achieved, is still within tolerance, or whether a warning or an alarm is present. From the length of a bar, the operator can determine how far the targets have been achieved, and if intervention is necessary.
Because of specific technical circumstances it was only possible to use a simplified version of SCARLETT. This new version of SCARLETT contains almost the same process elements as the original one, but malfunctions are realized in a different way. Leakage in the pipes can occur, breaks are not implemented.
4.1 The Virtual Process Elements Developed Each process contains a different set of process elements so that it is necessary to develop a special set of Virtual Process Elements. Based on previous projects (e.g. Wittenberg, 1998; Wittenberg, 1999), existing sets of process elements were extended to cover the new requirements.
4. APPLICATION
In addition to the extensions for the specific process, the functional properties were also enhanced. The new versions of Virtual Process Elements not only enable the current and desired process variables to be represented, they also indicate the direction in which the process variable should be changed. These features conform to the requirements for user-centered process-visualization (see above). These elements present the actual state, the optimal state and information about any necessary action, to assist the operator in performing his tasks.
At the Laboratory for Cognitive Systems Science (University of Tsukuba) an artificial microworld called SCARLETT (Supervisory ~ontrol ~nd Response to 1eaks: Iara at Isukuba) is used for experimental research on supervisory control (Inagaki et aI. , 1998; Moray et aI. , 2000). This microworld represents a central heating system to maintain the temperature of an apartment complex (Figure 2). It consists of three heatable reservoirs (R I, R2, R3), which are
83
green (optimal) through yellow (warning) to red (alarm). In addition, the usual symbols (e.g. a warning triangle) are displayed. The use of color and shape coding simplifies identification of the current state.
Figure 3 shows the visualization of the fill-up level in a reservoir. A small line represents the target value (50%). This line is visible only if there is a deviation from the target value. The arrows show the direction of the necessary level change. Like the "target-line" these arrows are again only displayed if there is a deviation. The size of each arrow corresponds to the deviation from the desired fill-up level.
=-iI._ Figure
This research project also encompassed the development of new visualization objects for the specific application example. For example, a visualization object was developed to show the temperature in the apartment complex (Figure 4). This object has the shape of a house (rectangle with roof). The color of this object is determined by the actual temperature. If there is a deviation from the desired temperature, a small colored line is visible. A triangle coverts this line to an arrow symbol, suggesting the necessary change oftemperature (up or down).
5: Visualization of target deviations (temperature of apartment complex) (left: actual T lower than desired T; center: actual T desired T; right: actual T higher than desired T)
Figure 3: Visualization of the actual and desired fill-up level (L) (left: actual L lower than desired L; center: actual L = desired L; right: actual L higher than desired L)
~~ D
__
Figure 6: Task-oriented interface based on the Virtual Process Visualization method (Task view: "Maintain temperature")
Figure 4: Visualization of the actual and desired temperature (T) in the apartment complex (left: actual T lower than desired T; center: actual T = desired T; right: actual T higher than desired T)
4.2 Visualization of deviations from target
qualitative
As a supplement to the objects displayed, the status of the process items is indicated by colored bars pointing in the direction of the deviation from target. If the bar is on the left-hand side (which is usually associated with a negative deviation) the process variable is too low. The state is indicted by color changes from
Figure 7: Task-oriented interface based on the Virtual Process Visualization method (Task view: "Maintain mass flow")
84
associated windows. Below the navigation buttons there are also indicators for the deviation of the actual process states from the targets. The operator can work in one task-oriented view but at the same time has an overview of all process views.
4.3 Task-oriented structure and the complete graphical user interface The main task is to control the temperature in the apartment complex to guarantee the supply of heat. Contributions to this objective are made by the reservoir heater, the heat exchanger, and pump P3.
5. EXPERIMENTS
The subordinate task is to maintain the mass flow. Without this mass flow, heat cannot be transported to the apartment complex. The mass flow is controlled by all the reservoirs, valves, pumps and pipes.
In the experiments, this visualization is be compared to and evaluated against a conventional topological interface. These experiments were intended to verifY the hypothesis that this visualization is more descriptive, thus making it easier to comprehend the actual process state.
Based on these two tasks, two process views have been defined. Figure 6 shows the view for the task "Maintain temperature in the apartment complex". This view contains mainly energybased process variables. All temperatures are color-coded, the flows through the pipes are related to the diameter of the colored pipes. The operator is given a clear view of the energy flow within the facility. The only target in this view is the desired temperature in the apartment complex.
The experiment was designed as a 2x2 matrix. Two different kinds of interface were used (Virtual Process Visualization vs. Conventional Topological Interface). The subjects (Japanese graduate students) were split into in two groups. Each group started with a different interface. After the first stage of the experiment, the type of interface was changed for the second stage. Each stage contained five different scenarios, i.e. different initial values and a different desired temperature in the apartment block.
The second task-oriented process view shows the The dependent variables were the mean errors of the task "Maintain mass flow" (Figure 7). This view temperature of the apartment block, the fill-up levels in the reservoirs, the time to respond to malfunctions contains all elements and variables which are and the number of undetected malfunctions. 30 associated with the mass flow. In this view there are further targets. The fill-up level in the different sets of variables were recorded. The variables were analyzed with the software tool reservoirs should be 50 %, leaving enough space to hold fluid flowing from preceding elements. STATISTICA. However, a fill-up level of 50 % also means that As expected, differences were found in the time to there is enough fluid stored to maintain the mass respond to malfunctions (Figure 8), the number of flow to the following process elements. Another undetected malfunctions (Figure 9) and in the mean target is associated with pump P3. This pump is error of the apartment block (Figure 10). An ANOV A located just after the heat exchanger, and it was carried out, and showed that all these results are determines the mass flow (and also the energy significant. flow) through the heat exchanger. Regardless of the current situation, there must be a flow through the heat exchanger. The flow Plot of Means (urlWelghtedl Interface Main Effec! through pump P3 must exceed a F( 1.22)=743. p<.0124 certain minimum rate. 35000,----------------------, Malfunctions like leakages are associated with the mass flow . For this reason, any loss of water is also indicated as a target deviation. Depending on the amount of the water loss, the operator can decide if there is a leakage or a break. In fact the simulation behind the interface is not able to represent breaks, but this limitation of the simulation is not known to the subject in the experiments (the operator).
-'-'
!
30000
E
25000
'0
c::
.g-
20000
.
15000
0
er 0
I-
E
;::
10000
5000
- !Conventional
Virtual
Interface
The top left area of both views Figure 8: Graph of mean values: time to respond to a malfunction includes a pane, with which the operator can navigate between the views. The navigation buttons are minimized views of the
85
There were no significant differences between the two types of interface in the mean errors of the fill-up levels_
Plot of Means (unwelghted) Interface Main Effect F(1.28)=7.00: p<.0132
07 0.6
A very interesting result was that the difference in the mean error of the temperature of the apartment complex depended on the sequence in which the interface types were used_The subjects who started with the interface based on the Virtual Process Visualization method made a significantly lower error (including when they used the conventional interface)_ Figure 11 shows a graph of these results_
Ti
0.5 0.4
i
0.3
i
0.2 0.1 00 -0.1 I
_L
-0.2 -0 3
Conventional
Vi rtual
Interface
Figure 9: Graph of mean values: undetected malfunctions 6. DISCUSSION
Plot of Means (unweighted) Interface Main Effect
F(1.28 )=6 43. p< 0171
~
3.0 , - - - - - - - - - - - - - - - - - - - - - - - - ,
"
2.8
"
~ 2.6
8 2.4
~ 2.2
-,,
t"
2.0
8.
I
1.8
I
'~"
E
1.6
,--
1.4
~ 1.2
i" t
L---------!
1.0 0.8
8 0.6 ~
04
~
0.2
i 1
I
~ 0.0' - - - - - - - - - - - - - - - - - - · ----Virtual
Conventi onal
Interface
Figure 10: Graph of mean values: mean error in the apartment complex temperature Plot of Means 2-way Interaction
F(1.28)=4 72: p<0385
u
~ 3.0 , - - - - - - - - - - - - - - -.-..
"
" 'ti.
-----,
2.8 2.6
~
2.4
~
2.2
~ 8.
2.0 1.8
:
1.6
~ 14
"~
1.2
~
0.8
~
0.4
"~
1.0
8 0.6
-i '·
t···
...............
.~ ~
_1 .•
c 0.2
~
....1... ..
c->v
0.0 ' - - - - - - - - - - - - - - - - - - - - ' Conventional
Vi rtual
ORDER
.. 0 .. ORDER ",,>c
Interface
Figure 11 : Graph of mean values: mean error in the apartment complex temperature as a function of the sequence in which the interfaces were used
86
In the experiment, two interesting points were found. The first is that there are no significant differences in the mean errors of the fill-up levels of the reservoirs. In discussion with the subjects it emerged that subjects paid the greatest attention to controlling the temperature in the apartment complex. This suggests that the hierarchy of targets was presented clearly. The other interesting fact is that there is a difference depending on the sequence in which the two interface types were used. Subjects who started with the interface based on the Virtual Process Visualization method made significantly less errors than the subjects who started with the conventional interface. This result could prove the hypothesis that a highly pictorial interface presents the process knowledge more clearly and assists in the building of a mental model (see above). Using the Virtual Process Visualization method, the technical is presented more system transparently than with the conventional technology-oriented method. The other results (mean error temperature, time to respond of to malfunctions, number undetected malfunctions) also confirm the effectiveness of this visualization concept.
Wittenberg, C (200 I): Virtuelle Prozessvisualisierung am Beispiel eines verfahrenstechnischen Prozesses. VDI-Fortschrittsbericht Reihe 22 Mensch-Maschine-Systeme Nr. 5, Diisseldorf: VDI-Verlag. ACKNOWLEDGEMENT
REFERENCES DIN 28004 Teil 3 (1988): FlieBbilder verfahrenstechnischer Anlagen; Graphische Symbole. Dutke, S. (1994): Mentale Modelle: Konstrukte des Wissens und Verstehens. Gottingen: Verlag flir angewandte Psychologie. Engelkamp, J. (1991): Das menschliche Gedlichtnis: Das Erinnern von Sprache, Bildern und Handlungen. 2. Aufiage, Gottingen: Hogrefe. Inagaki, T., N. Moray, M. Itoh (1998): Trust, Self-Confidence and Authority in HumanMachine Systems. In: Proceedings IF AC Man-Machine Systems, Kyoto, pp. 431-436. Johannsen, G. (1993): Mensch-MaschineSysteme. Berlin: Springer-Verlag. Moray, N., T. Inagaki, M. Itoh (2000): Adaptive Automation, Trust, and Self-Confidence in Fault Management of Time-Critical Tasks. Journal of Experimental Psychology, 2000, Vol. 6, No. 1, pp 44-58. Norman, D.A. (1986): Cognitive Engineering. In D.A. Norman, S.W. Draper (Eds.): User centered System Design, Hillsdale New York: Earlbaum, pp. 31-62. Rasmussen, J. (1984): Strategies for state identification and diagnosis in supervisory control tasks, and design of computer-based support systems. In: W.B. Rouse (Ed.): Advances in Man-Machine Systems Research, Volume 1, Greenwich: JAI Press, pp. 139193. Wittenberg, C. (1997): Unterstiltzung der menschlichen Informationsaufnahme durch ProzeBvisualisierung mittels virtueller ProzeBelemente. In: K.P. Glirtner (Ed.): Menschliche Zuverllissigkeit, Beanspruchung und benutzerzentrierte Automatisierung, Bonn: Deutsche Gesellschaft fur Luft- und Raumfahrt e.V. (DGLR), pp. 183-196. Wittenberg, C. (1998): Qualitative visualisation of a chemical process using the threeIn: dimensional computer graphics. Proceedings of the 17th European Annual Conference on Human Decision Making and Manual Control, France, Valenciennes: LAMIH Laboratoire d' Automatique et de Mecanique Industrielles et Humaines, Universite de Valenciennes, pp. 217-226. Wittenberg, C. (1999): Aufgabenorientierte Visualisierung eines komplexen verfahrenstechnischen Prozesses unter Verwendung dreidimensionaler Computergrafik. In: U. Arend, E. Eberleh, K. Pitschke (Eds.): Software-Ergonomie '99 Design von Informationswelten, Berichte des German Chapter of the ACM Nr. 53. Stuttgart: B.G . Teubner, pp. 335-344.
This project was made possible by the support of the Japanese Ministry of Science, Education, Sports, and Culture MONBU-SHO and the Laboratory for Cognitive System Science of Prof. T. Inagaki (University of Tsukuba, Japan).
87