Application of Induction Techniques to the Monitoring of Large Dams

The present paper describes an application of machine learning techniques to structural engineering. The existence and availability of a large compilation of data describing some 1100 cases of structural damage in large dams, has contributed with a specific domain of application. The work has progressed in three different lines: 1) experimentation with existing induction systems; 2) identification of classification criteria for given domains in order to establish which problems are addressable by which algorithms; 3) implementation of a flexible induction tool enabling experimentation with disparate types of data and with different types of induction algorithms. INTRODUCTION The potential for machine learning systems in structural engineering has been prompted and illustrated by several authors in the past (Arciszewski 1989), but does not seem to emerge as a main concern for many of the groups involved in AI for structural engineering. The work herein described reflects a pragmatic approach to the area of machine learning with a view to its application in structural engineering domains. The existence and availability of a large compilation of data describing some 1100 cases of structural damage in large dams (ICOLD 1981), provided a domain of application able to close the above mentioned motivation. The expected result of applying machine learning techniques to those reports of dam accidents and incidents, was the identification of rules explaining given patterns of damage or failure. BASIC CONCEPTS Induction is also known as learning from examples. In Artificial Intelligence, to induct means to build (to learn) a classification from a set of given examples. Several techniques may be used 1 Assistant Lecturer 2 Assistant Professor for induction, according to a division in two main groups (Rich and Knight 1983): Induction Techniques Classes described statistically (by scoring functions) Technique of coeficient adjustment.    Classes described structuraly Learning guided by a teacher. Production of decision rules. Production of decision trees.   