Neural-network enhancement for a reliability expert-system

An inexpensive diagnostic system called Ride Expert recognizing rough ride problems and their causes in heavy trucks has been designed and built using expert system architecture. Two sources of input information were utilized in this system: a multiprobe vibration analysis system (MVAS) and a driver and/or technician from truck services. It was soon apparent a classic expert system approach was impractical due to the enormous number of inference rules necessary to draw diagnostic conclusions. To reduce this requirement we added a neural network trained to analyze data acquired from MVAS. It greatly improved Ride Expert's robustness and reduced the number of necessary rules to about 1/3. This paper presents the Ride Expert development stages and what we have learned from this process.<>