Expert System for Sewer Network Maintenance: Validation Issues

Abstract

A program to improve the sewer maintenance activities in a suburb of Paris, France, includes an expert system to diagnose the need for sewer repairs. Organizational constraints (e.g., insufficient funds for knowledge engineering and difficulties accessing required computer facilities) precluded use of standard rapid prototyping development procedures. Validation tests were conducted to investigate whether the system's outputs could be used with confidence. An internal validation was performed to see if, on each of two test case sewer sections, the expert system would yield the same diagnoses as the geotechnical experts who formulated the rule trees. An external validation was performed on three additional test case sewer sections to compare the system's diagnoses with those of geotechnical specialists who were not involved in developing the knowledge base. The validation exercises demonstrated drawbacks of not using rapid prototyping development methods. Results included recommendations to make internal validation an integral part of knowledge acquisition, a design for a full‐scale validation of a more complete version of the system, and indications of how future validation studies can be made more efficient.