Unsupervised environmental operating condition compensation strategies in a guided ultrasonic wave monitoring system: evaluation and comparison

Abstract

Guided ultrasonic wave (GUW) monitoring systems are gaining much attention in pipeline condition monitoring. However, the effects of environmental and operational conditions (EOCs), especially temperature and random noise, degrade damage detection performance. When EOC effects produce greater amplitudes than the reflected waves from small damage cases, the reflected waves remain unidentified. This paper proposes an unsupervised learning-based denoising autoencoder (DAE) to reduce the effect of EOCs in GUW monitoring systems. A DAE decodes high-dimensional data into low-dimensional features and reconstructs the original data from these low-dimensional features. By providing GUW signals at a reference temperature, this structure forces the DAE to learn the essential features hidden within complex data. The proposed DAE undergoes comparative analysis with other popular unsupervised learning algorithms used for EOC compensation in GUW monitoring systems, such as principal component analysis, independent component analysis and deep autoencoder algorithms. EOC compensation performance is evaluated through receiver operating characteristics (ROC). From the numerical model and an experimental model, the GUW database is obtained. All four algorithms showed good damage detection performance using a numerical model; however, in the experimental model, the proposed DAE showed superiority among other methods.