A practical application of probabilistic neural networks to machinery failure prevention

摘要

One of the most difficult tasks in building expert systems, and real-time diagnostic systems in particular, is dealing with the issue of uncertainty. Uncertainty arises in several places, the most common ones being in sensor measurement and in abnormal symptom (alarm) generation. Sensor failure, which includes out-of-calibration condition, is a common occurrence aborard ships, yet all high level automation functions, such as monitoring and control, performance analysis, and diagnostics/prognostics, critically depend on accurate sensor measuement inputs. Uncertainty also arises in the classification of the sensor inputs into nonmal and abnormal states. Uncertainty related to sensor failures and alarm genration can result in erroneous and/or unreliable performance of diagnostic system.