|
   |
||||||||
|
|||||||||
|
|||||||||
TTL/PIC4SAT |
MoDyS-Project TTL [aka PIC4SAT] Towards a Theory of Trend LiteracyTrend views show the history of a process as time-ordered curves of selected measurement data. In particular during dynamic processes like startup or shutdown they are of utmost importance for the operator’s process-understanding and interpretation. Some interesting applications in the field of process monitoring and error diagnosis claim to adopt human trend-reading principles. The basic idea is to transfer the time-ordered sequence of crisp process data to a sequence of symbols describing the overall course of the process. Results of this research are used within the project MoDyS TLA - (Trend Literacy Applied ).
Trend literacy - the ability to perceive and interpret trend charts according to the present situation - seems to be of utmost importance for the operators' process-understanding and their situation awareness of the processes in control. Various applications in the field of process monitoring and error diagnosis claim to adopt human trend-reading principles. The basic idea is to transfer the time-ordered sequence of crisp process data to a sequence of symbols describing the overall course of the process. The sequence is divided into segments of fixed or varying length, that then are classified to be a member of a set of basic building blocks (trendlets) e.g. by means of artificial neuronal nets (Vedam & Venkatasubramanian, 1997), or by selecting temporal fuzzy logic templates (Haimowitz & Kohane, 1993). The application of these algorithms in an operator training system, that addresses the concept of situation awareness (Endsley, 1995), seems to be very promising. The situation specific relevance of elements and constellations of elements is computed by a cognitive model that reflects an expert operator's reasoning. The systems "conclusions" are brought to the trainee's attention e.g. by highlighting (Leuchter & Urbas, 2002). To quantify particular aspects of the expert model, some questions need to be answered: How do human operators divide trend charts into segments and how are these segments classified? Do they rely on perception based pattern matching only, or is extant process knowledge utilized to interpret the signals? How stable are these findings in relation to the research methods applied? Can one obtain the same effects in prediction task? How does the presence of a predefined set of trendlets affect the segmentation decision? Does this predefined set correspond to categories within the human cognitive system? Do process plant operators all use the same set of categories or do they depend on the operators' specific experience? These questions are addressed in a series of three experiments and one field study. The experiments were carried out as laboratory experiments using a web based experiment control system based on the MoJavEE-framework (Kindsmüller, Leuchter & Urbas, 2003). Participants of all three experiments where graduate students of process engineering from the Department of Process Dynamics and Operation at Berlin University of Technology. The field study included process plant operators from six sites of three large companies within the field of chemical base products production (Henkel/Cognis, Degussa, Bayernoil). Results reveal that subjects not only take advantage of the induced knowledge in higher cognitive tasks like categorization, but are affected by this knowledge in more perception based tasks as well: e.g. knowledge of the trend generating process has a strong impact on the precision of operators separation decisions. This suggests that the interpretation of trend diagrams is influenced by strong top-down components that have to be considered for modeling the operators' expertise in trend reading.
|
||||||||||||
 |
© 2000-2005 by MoDyS-Team, ZMMS, TU Berlin | Impressum | Disclaimer | |
| mailto: modellierung (at) zmms.tu-berlin.de last modified: Dezember 20, 2006 15:22 |