Knowledge-based improvement: simulation and artificial intelligence for identifying and improving human decision-making in an operations system

Abstract

The performance of most operations systems is significantly affected by the interaction of human decision-makers. A methodology, based on the use of visual interactive simulation (VIS) and artificial intelligence (AI), is described that aims to identify and improve human decision-making in operations systems. The methodology, known as 'knowledge based improvement' (KBI), elicits knowledge from a decision-maker via a VIS and then uses AI methods to represent the decision-making. By linking the VIS and AI representation it is possible to predict the performance of the operations system under different decision-making strategies and to search for improved strategies. The KBI methodology is applied to the decision-making surrounding unplanned maintenance operations at a Ford Motor Company engine assembly plant.

Simulation based knowledge elicitation: Effect of visual representation and model parameters

“…Anecdotal evidence of using simulation to elicit expert knowledge is found in Flitman and Hurrion (1987), Shaw (1989), Pierreval and Ralambondrainy (1990), Chryssolouris et al (1991), Hurrion (1991), Liang et al (1992), Tan et al (2000), Tan (2003) and Robinson et al (2005). These are a few examples of what Ören (1994) classifies as using simulation for applications of AI.…”

“…However, the original plan to learn human behavioural rules was not carried out. Robinson et al (2005) develop the knowledge-based improvement methodology which uses VIS to elicit knowledge from decision-makers; in this case maintenance supervisors in a manufacturing plant. The data collected are then used to train an AI model which is subsequently linked to the VIS model to represent the human decision-makers.…”

“…It is not, however, without its limitations. Robinson et al (2005) identify three specific problems with using VIS for knowledge elicitation:…”

“…The benefit of presenting the expert with more uncommon and extreme scenes is that there is less probability of repeating the same scene many times, leading to expert boredom and fatigue (Robinson et al, 2005). The expert is also more likely to find more occasions on which to intervene and interact with the model.…”

“…A particular issue arises when there are no consequences from making poor decisions in the model (Robinson et al, 2005). To be more effective in knowledge elicitation, then incentives in the model need to be aligned more closely with the real system.…”