Presentation is loading. Please wait.

Presentation is loading. Please wait.

Problem solving in control of discrete-event systems Lenko Grigorov and Karen Rudie Queen’s University Kingston, Canada.

Similar presentations


Presentation on theme: "Problem solving in control of discrete-event systems Lenko Grigorov and Karen Rudie Queen’s University Kingston, Canada."— Presentation transcript:

1 Problem solving in control of discrete-event systems Lenko Grigorov and Karen Rudie Queen’s University Kingston, Canada

2 July, 2007Grigorov & Rudie, Queen's Univ.2 Content Motivation Observational study Data analysis methodology Results and discussion Future directions

3 July, 2007Grigorov & Rudie, Queen's Univ.3 The look of DES software

4 July, 2007Grigorov & Rudie, Queen's Univ.4 Problems with DES software No facilities to represent huge models meaningfully (10 6 + states) Does not support much besides performing DES algorithms Formalizing an informal model Verifying the output of algorithms Implementing supervisors

5 July, 2007Grigorov & Rudie, Queen's Univ.5 How to address the problems? The problems with DES software are complex No straight-forward solution Study done by Rogers et al. on diagnosis from X-rays Understand cognitive processes Use information to design software interface

6 July, 2007Grigorov & Rudie, Queen's Univ.6 Goal Understand human problem-solving strategy in control of DES Create a model of the cognitive process Use the model to guide the development of DES software Test the new software to validate improvements

7 July, 2007Grigorov & Rudie, Queen's Univ.7 Observational study 5 experts asked to solve DES problems Definition of problem: informal description Expected solution: formal model and DES supervisor(s) Use pen and paper and/or software Switch as many times as desired Verbalize thinking Performance recorded with video camera

8 July, 2007Grigorov & Rudie, Queen's Univ.8 Analysis approach 1.Use prior experience to create taxonomy of DES problem solving 2.Refine taxonomy from observed data 3.Encode data according to taxonomy 4.Analyse encoded form to find patterns [as per Rogers et. al.]

9 July, 2007Grigorov & Rudie, Queen's Univ.9 Data encoding Encoding activities along 4 main axes Type of activity Perform with pen and paper, perform with computer, verbalize... DES entity referred to Module, event, state... Stage Inspection, verification... Action Create, modify appearance, count...

10 July, 2007Grigorov & Rudie, Queen's Univ.10 Data analysis – application One video session encoded and analysed Two periods Pen and paper Computer Duration of activities N-gram analysis and clustering

11 July, 2007Grigorov & Rudie, Queen's Univ.11 Data analysis – n-grams N-gram analysis: the ratio of occurrence of a specific sub-sequence of n items in a larger sequence Sequence 'abcdbbc', 2-gram 'bc' Absolute ratio is 2/6 Relative ratio is 2/3 Relative to all n-grams which start with the same (n-1) symbols

12 July, 2007Grigorov & Rudie, Queen's Univ.12 Data analysis – clustering Unsupervised clustering: assign data items to separate classes No prior idea of How many classes What the criterion of distinction is Distance between items is bigger if Type of item is different Time between items is larger

13 July, 2007Grigorov & Rudie, Queen's Univ.13 Analysis of references Reference to entities DES modules FSA elements: states, transitions, events Computational algorithms Reference to DES modules Machines 1 and 2 Buffers 1 and 2 Testing unit

14 July, 2007Grigorov & Rudie, Queen's Univ.14 Output of N-gram analysis

15 July, 2007Grigorov & Rudie, Queen's Univ.15 Output of clustering

16 July, 2007Grigorov & Rudie, Queen's Univ.16 Preliminary results (1) 12 min pen and paper 7 min reading and understanding problem Rest for modeling 34 min computer 8 min (23%) improving layout of graphs Rest for input of models, DES algorithms, verification and remodeling

17 July, 2007Grigorov & Rudie, Queen's Univ.17 Preliminary results (2) Subject works with “chunks” of related activities Type of entity: if working on states, not likely to interrupt with work on events Module: if working on machine1, not likely to interrupt with work on machine2

18 July, 2007Grigorov & Rudie, Queen's Univ.18 Preliminary results (3) Subject does not consider DES algorithms if thinking at the low level of states, transitions, etc. Only when thinking at the level of modules Software seems to shape workflow Pen and paper: no predominant pattern Computer: modeling in the sequence “module, events, states, transitions” This is the sequence supported by the software

19 July, 2007Grigorov & Rudie, Queen's Univ.19 Discussion Discrepancies between the two periods Different stages of problem solving Software imposes constraints Graphical representation of model is very important Software not suitable for conceptual modeling Subject chose pen & paper in the beginning

20 July, 2007Grigorov & Rudie, Queen's Univ.20 Current research - conceptual modeling In the initial stages of design, subjects Consider participants/ sub-systems and Interactions between them

21 July, 2007Grigorov & Rudie, Queen's Univ.21 Current research - framework for conceptual modeling Template design of DESs Inspired by observations Library with templates of common behaviors Instantiate templates Link them No need to consider low- level details

22 July, 2007Grigorov & Rudie, Queen's Univ.22 Future work Analyse all video sessions Improve encoding scheme Use other analysis techniques Build model of problem-solving strategy What steps are taken What information is needed and when Use model to improve DES software

23 July, 2007Grigorov & Rudie, Queen's Univ.23 Queen’s University


Download ppt "Problem solving in control of discrete-event systems Lenko Grigorov and Karen Rudie Queen’s University Kingston, Canada."

Similar presentations


Ads by Google