Machine Learning Algorithms for Health Monitoring of Timber Utility Poles Using Stress Wave Propagation

Abstract

Stress wave propagation (SWP) technique is a simple and cost-effective non-destructive testing technique which can be effectively employed for the health monitoring of timber utility poles. In this paper, Artificial Neural Network (ANN) pattern recognition algorithm is used for the classification of stress wave responses obtained from testing in-service timber poles. Thirty in-service timber poles in Victoria, Australia are tested which belong to different timber species and varying geometric parameters. The tested poles are uprooted and subjected to full scale bending tests in order to determine the failure moments. Health status of each pole is defined based on the ratio between the failure moment and the design moment capacity. 252 stress wave responses are obtained from the field testings by the application of different impacts. An ANN model is developed to classify these signals based on the defined target groups according to the health status. The mobility spectrum of the pole responses in the low frequency region and the pole diameters are selected as the inputs to the ANN model. The performance of the developed ANN model is evaluated by calculating some performance parameters. Further, Support Vector Machine (SVM) and k-nearest neighbors (k-NN) algorithms are also applied to the same data set for classification. The performance of each technique is compared to select the best performing method. Results of this study showed that the developed ANN model outperforms the other techniques for the condition assessment of timber poles using the stress wave propagation technique.