Use of Artificial Neural Networks for Automatic Data Plausibility Check and Test Data Quality Improvement

摘要

The objective of this work is to explore new methods and algorithms to automatically check, assess andrnimprove the quality of data acquired through structural testing.rnWhen dealing with large amounts of data, one of the main problems coming from intensive measurementrnsessions is the validity and consistency of the data set. Apart from the possibility to set fixed ranges forrnmeasurement channels and measure over/underloads, users do not have many tools for checking the plausibilityrnof the data in an automated way. Furthermore, envisaged criteria for the acceptability of the data depend on thernobject under analysis, the test environment and the environmental conditions. In structural testing one of thernmost usually applied criterion is the check on the FRFs with special concern to the driving point. This crossrncheck is quite time consuming and it requires user interaction.rnThis paper addresses the problem of automatically assessing the quality of data measured during intensivernstructural testing.rnAn automated procedure based on Artificial Intelligence is presented to evaluate data plausibility. ArtificialrnNeural Networks (ANN) are in use for the analysis of Frequency Response Functions (FRFs) taking advantagernof their pattern recognition capability. ANN technology provides automated interpretation of data quality,rnallows cluster analysis and blind separation of faulty data.rnIn particular Multi Layer Perceptron (MLP) networks are used to automatically recognize faulty FRFs, basedrnon signal properties as the noise/signal ratio and phase.