Technical University of Crete
Dept. of Production Engineering &
Management, Crete, Greece
Recent Developments in Task Modelling
Tom Kontogiannis
1
Operational Problems Giving Rise to
Hazards & Incidents in Process Industries
• Unanticipated events that require modifications in
the procedures & methods required to do the job
• High workload experienced by human operators
• Changing the allocation of tasks and taking over
new responsibilities without adequate practice
• Coping with interruptions and suspended tasks
• Changing plans & priorities under time pressure
• Recovering erroneous actions that occurred
2
Use of Task Analysis Methodologies
in the Control of Major Hazards
• Task analysis methodologies are used to examine:
– Goals and plans utilized by operators to perform tasks
– The allocation of tasks between different operators
– Critical task information obtained from control panels
and operating procedures
• They can give an indication of operator workload
• They provide input to human error analysis
• Their output is used to optimize procedures, job
methods, control panel design, and training
3
Task Analysis Methodologies
•
•
•
•
•
Hierarchical Task Analysis
Operational Sequence Diagrams
Link Analysis
Time Line Analysis
Cognitive Task Analysis
(Ainsworh & Kirwan, 1990)
4
An example of Hierarchical Task Analysis
PLAN 1. At all times assess situation. Assign staff
& tools first and continue with plan implementation
Perform task
PLAN 1
Assess situation &
decide goal sequence
Assign
staff & tools
PLAN 5. Do T6 then T7 then do T89 & T1011 together
If workload is high abort task or assign to other person
PLAN2 (assess situation). Start with goal 3.
If temp=high, do goal 1 aborting rest.
If temp=medium, do goal 3 providing goal 1
is not in progress. Finish with goal 2
“Assess situation”
INPUTS
“Process goals”
Attend to
pending tasks
Task 89
PLAN 9. Choose task with lower workload
PLAN 4
Perform goal 2
Plan 5
Task 7
PLAN 8 & 10. Do in parallel. If workload is high
do in serial or assign task to other person
Release
staff & tools
Perform goal 1
Task 6
PLAN 7. Do all in sequence. If workload is high
then abort task or assign to other person
PLAN 3
Do in sequence
Process
goals
(i.e., due to error
or high workload)
PLAN 4. Select goals according to the planned sequence
of goals of the assessment process
Implement/Plan
sequence of goals
Perform goal 3
PLAN 6
Do in sequence
PLAN 7
Task 1011
Task 1
PLAN 8
Task 2
Task 3
Task 45
PLAN 9
PLAN 10
Task 8
Task 9
Task 10
Task 11
Task 12
Task 13
Task 4
Task 5
5
Inadequacies of Task Analysis
• Task Analysis describes “how the job should be
done” under a set of well-defined job conditions
• T.A. runs into difficulties in cases where system
events require modifications in goal priories, task
sequences, and allocation of tasks
• It is difficult to model unanticipated events
resulting in interruptions and suspended tasks
• T.A. cannot model adequately the workload
encountered by operators
6
From Task Analysis
to Task Modelling
• Task modelling is a new set of methodologies that
provide computer simulations of human tasks
• The output of task analysis (e.g., task sequences)
can be used in specifying a task network
• A task model represents the control of tasks, the
flow of information, & the utilization of resources
• Software tools can be used to verify the task
model under a wide range of conditions (e.g.,
stochastic duration, external events, human errors)
7
Task Modelling Techniques
•
•
•
•
•
GOMS (John & Kieras, 1996)
MicroSAINT (Laughery & Corker, 1997)
ConcurTaskTrees (Paterno et al, 1997)
TOBOLA (Uhr, 2002)
DIANE+ (Tarby & Barthet, 1996)
8
Data Requirements for
Specifying Task Networks
• Task accuracy
• Task duration
(prob. distribution)
• Task release condition
• Beginning effect
• Ending effect
• Primary/contingent
operators
•
•
•
•
Task priorities
Task loadings
Probability of failure
Failure consequences
in time and accuracy
• Branching conditions
(e.g., probabilistic,
tactical, multiple)
9
Output of Task Modelling
•
•
•
•
A time line of completed tasks
A record of interrupted, resumed or failed tasks
The allocation of tasks to operators
An estimate of operator workload (e.g., analogous
to the number of tasks carried out in parallel)
• Performance measures (e.g., speed & accuracy)
• An evaluation of the consequences of task failures
10
Inadequacies of Task Modelling
• Utility problems
– Exhaustive data requirements
– Expert estimates on data (e.g., task loadings)
– Expertise in task modelling languages and codes
• Functional problems
– The lack of a USER MODEL to indicate how operators
process task information and re-schedule tasks
– The lack of modelling several diagnostic and decisionmaking activities
11
A New Approach to Task Modelling
based on Petri Nets
• TASK MODEL
– Petri Net graphs are used to build the task network
– A taxonomy of 15 types of task sequences (plans)
– Task templates regulate information to the User Model
• USER MODEL (expressed in mathematical terms)
– Recall-Forget model, Task Selection model, Operator
Assignment model, Human Error Execution model
• DIAGNOSIS & DECISION MAKING
12
A New Approach to Task Modelling
MONITORING
& DIAGNOSIS
DECISION
MAKING
USER MODEL
TASK MODEL
Goal diagram
Recall tasks
TASK
TEMPLATE
Select tasks
Task diagram
Assign roles
Execute tasks
13
A Triple Representation of Task Networks
• The Petri Net Graph helps to visualize task
relationships and interactions
• The Mathematical modelling of the network helps
the application of formal analysis techniques
• The code segments of transitions helps the
programming of high-level functions and routines
(e.g., implementation of user models)
14
A Coloured Petri Net (CPN)
Representation of Task Networks
• A Net graph consisting of nodes and arcs
• Places hold information about resources, releasing
conditions, beginning and ending effects
• Transitions represent tasks that manipulate the
information held in adjacent places
• Tokens are ‘data items’ activating places as they
move around them
• Arcs are inscribed with expressions (e.g.,
functions, variables)
15
Mathematical notation of Petri Nets
• A Petri Net is a birartite directed graph
– G = [ P, T, A ] where:
• P is a set of places
• T is a set of transitions
• A is a set of directed arcs
• The state of the system is represented as a matrix
of marked places and enabled transitions
• Formal analysis techniques can be used
on the task network
16
Start
Goal 1
Goal 3
Goal 2
The Task Model as
a Petri Net Graph
T1
T2 T7
T8 T12
T3
T9
T14
T6
T11
T15
End
T13
17
SEQUENTIAL PLANS
DISCRETIONARY PLANS
Fixed sequences: Do A and B in specified order
Discretionary inclusive plans:
Do both A and B, in any order, or concurrently
A
B
A
T
A Taxonomy of Plans
Prioritized sequences: Do both A and B
giving priority to A
A
B
Discretionary exclusive plans:
Do either A or B, in any order
secondary
place
A
secondary
place
B
Unordered sequences: Do both A and B
in any order, but not concurrently
A
B
Optional plans: Do either A or B, or both
A
T
secondary
place
B
[* Guard is to prevent transition T from firing twice]
B
Interleaved sequences: Start B before A is completed
pause
A1
N out of M joins: Do only n out of m tasks
A
T1
B
T2
C
T3
A2
@ + delay
secondary place
B
[guard]
18
A Task Template for passing Information
between Task and User Models
Tin_2
Task
t
(g,id)
Int
Task_ID
[t=id]
t
t
g
Data
Tin_1
USER
MODEL
Super_in
BEGIN
g
tsl
TaskStateList
AllStates
Data
Super_out
tsl
If s<>Done &
S <> Started
then 1`(g,t)
else empty
s1
(g,turn,t_out,imp)
1` Default
s2
State
State
s1
s2
END
[t=t_out]
If pass=on
then 1`g
i+1 else empty
T_out
i
Counter
If s<>Done
then 1`g
else empty
19
The User Model
Super_in
(g,t)
Super_mid
ATTEND
TASKS
(g,t,s)
Selected
tasks
Super_out
(g,t,s)
PERFORM
TASKS
(g,t,s,imp)
stm
stm
STM
Tasks
Tasks to
be recalled
(g,t)
stm^^[(g,t)]
20
The Human Memory and
Selective Attention Models
RecDataList
TaskStateList
rdl
rdl
PrioDataList
tsl1
tsl1
tsl2
tsl2
Super_in
pdl
pdl
Super_in
(g,t,s1)
ShedConsList
scl
scl
RECALL
TASKS
(g,t)
If s<>Default then
1`(g,t,s,p,a,dur)
else empty @+ delay1
msl1
If pass=on
then 1`(g,t)
else empty
ManStateList
BodyList
bl
bl
STM
stm2
(g_out,
t_out,
Started)
msl1
msl2
msl2
stm1
Super_mid
SELECT
TASKS
stm1
stm2
21
Operator Assignment, Workload & Human Error
FailConsList
fcl
fcl
fcl
fcl
TaskStateList
tsl1
tsl2
tsl1
tsl2
ManStateList
msl1
msl1
msl2
msl2
Super_out
Super_mid
(g,t,s1)
TaskXPerson
ASSIGN
ROLES
(g,t, Default)
Active
task
list
If s<>Default then
1`(g,t,s,p,a,dur)
else empty @+ delay1
tcl
(g,t)
(g,t,s,p,a,dur)
TaskCharList
Task
tcl
tcl
tcl
Task
EXECUTE
TASKS
ManCharList
Do in serial
If s=Default
then 1`(g,t) else
empty @+delay2
mcl
mcl
mcl
mcl
ConstraintList
cl
cl
cl
cl
22
Declaration of Colours (1/2)
• TaskCharList contains task loadings and conflicts on three
resource channels (i.e., perception-choice-action)
• TaskStateList contains an updated record of starting and
finishing times, assigned operators, status of tasks,
priorities of tasks, and state of goals (switches)
• ManCharList contains task information about durations
and accuracies for different operators for each task
• ManStateList contains an updated record of the workload
and availability of operators
23
Declaration of Colours (2/2)
• RecDataList contains data about cost of forgetting, and
strength of reminding cues for each task
• PrioDataList contains data about cost of task interruption,
deadlines to avoid consequences and cost of consequences
incurred
• ConstraintList contains data about the allocation rules that
should be applied to each task
• FailConsList contains data about side effects to tasks
upstream or downstream a specific task
24
A Model of Operator Assignment
Select operators with assignment scores
above a limiting condition
n
R = P (A/D)  Cj
j=1
R: Assignment score for a specific operator
P: priority of a specific task
A: task accuracy for an operator
D: task duration for an operator
j= 1 … n, rules for task allocation
Cj: operator compliance with each allocation rule
25
A Model of Operator Workload
(Multiple Resource Theory)
n
m
n-1
n
WT = t=1 ( Lt ) + i=1 ci t=1 s=t+1 (a t,i + a s,i )
WT: instaneous workload at time T
i= 1 … m are the resource channels (e.g., perception)
t,s= 1 … n are the operator tasks
Lt: standard loading for task t
a t,i : loading on channel i to perform task t
a s,i : loading on channel i to perform task s
c i = conflict between tasks sharing channel i
26
Strategies for Workload Management
•
•
•
•
•
•
A penalty is incurred in terms of accuracy & speed
Do not begin next task (Defer or skip next task)
Execute tasks sequentially instead of parallel
Interrupt ongoing task in favour of new (Replace)
Reassign ongoing task to another operator
Allocate new task to a less busy operator
27
A Model of Execution Errors
(Errors are related to high workload)
•
•
•
•
No error impact – next tasks is attended
Task is mistimed (premature or delayed start)
Next task cannot be started
Side effects are incurred, having an impact on previously
completed tasks
• Immediate error impact in terms of a penalty in task
duration, accuracy and task loading
• Error impact occurs when unsuccessful task is repeated
(e.g., operator assignment may change)
• Commission errors are modeled as separate task networks
28
Output of the Methodology
•
•
•
•
•
•
•
•
A time line of completed tasks
A graph showing operator workload at all times
A record of tasks forgotten due to system changes
A record of interrupted, resumed, repeated tasks
A record of changes in goal priorities
A record of operator assignment scores for tasks
Performance measures (e.g., speed & accuracy)
An evaluation of human error consequences
29
Advantages of the Methodology
• Takes into account the CONTEXT OF TASKS
(e.g., current workload, interrupted tasks)
• Workload takes into account mental activities
(e.g., diagnosis) in addition to task loadings
• Different USER MODELS can be tested
(e.g., memory model, selective attention model)
• The relationship between high workload and
human error is better explored
• Diagnosis and task planning are integrated
30
Disadvantages of the Methodology
• Exhaustive data requirements, especially with the
addition of user models
• Alternative plans must be thought of and specified
in advance
• Commission errors are modeled as separate task
networks and this may clutter the diagrams
• Reliance on availability of human error
probabilities
31
Future Developments
• Verification of human memory and selective
attention models in the context of a real scenario
• Development of a model for diagnosis and
decision making
• Development of an architecture for ‘abstract’
planning to be integrated with the task network
32
Descargar

Document