An intelligent system approach for construction risk identification

This dissertation presents a unique construction planning tool based on AI principles, decision analysis formalism, and real problem data. This tool provides a mechanism for feedback problem experiences into the identification, analysis, and response to risks for an upcoming project. It also furnishes modeling and formal analytical methods to model and explore probabilistic implications for novel and uncertain situations by incorporating decision-makers' own judgments with the knowledge acquired from multiple experts. Construction Risk Identification System (CRIS), the prototype system developed in this research, consists of a risk identification knowledge base, an inference engine, a knowledge acquisition subsystem, an explanation subsystem, and graphic functions that are implemented to enhance the man-computer interaction for each subsystem. Current available information in the knowledge base is acquired from 212 interviews and questionnaires, as well as, an extensive review of existing publications. Using database management system functionality, each piece of this knowledge is explicitly modeled according to its project characteristics, hierarchical taxonomy, and causal relations. This explicitly modeled knowledge is explainable and accessible by many application programs. The methodology involved in developing the inference engine is outlined as follows. Probability axioms are used to formalize the influence diagramming approach as an unambiguous language for representing uncertainties in construction projects. Bayes theorem is applied in conjunction with the influence diagramming technique to serve as a reasoning framework for project risk models. Practical and theoretical issues in combining multiple experts' assessments are discussed. For computational and communicational purposes, combining methods are also proposed to aggregate and represent multiple experts' judgments into a single distribution. A simulation technique is then applied to analyze risk models represented by influence diagrams. An iteration algorithm is developed to: (1) automatically construct influence diagrams using the knowledge base, (2) detect probabilistic independences embedded in influence diagrams, (3) examine completeness of probability assessments in influence diagrams, (4) elicit the minimum required information to analyze influence diagrams, and (5) automatically generate local computation formulas for the simulation method. A knowledge acquisition subsystem is developed to reduce documentation efforts required from experts and control the consistency of the knowledge base as it expands. An explanation subsystem is also provided to help experts and users in comprehending the line of reasoning and results generated by the system. Finally, a consultation example is included to demonstrate how a user might use this system to explore implications of alternate management and information gathering actions.