Application of MARSplines Method for Failure Rate Prediction

Abstract

In this paper MARSplines method was presented to model failure rate of water pipes in years 2015-2016 in the selected Polish city. The output parameters were chosen as three dependent variables - three values of failure rate of water mains, distribution pipes and house connections. Diameter, season, material and kind of the conduit were selected as independent variables. At the beginning of modelling 21 basis (splines) function were assumed. On a final note two functions were selected (after reduction of negligible functions). The model consists of three factors: β0, β1 and β2. The penalty for adding basis function was assumed at the level of 2. The correlation was equalled to 0.44. Relatively huge discrepancies between real and predicted values of failure rate of water mains and house connections were observed. In the future investigations concerning this problem the three separated models for each kind of conduit should be created. The calculations using MARSplines method were carried out in the program Statistica 13.1.

MARSplines method as a tool for failure frequency modelling

“…It means that MARSplines method could not be recommended as regression algorithm for forecasting of failure frequency of water pipes. Even building two separate models for each kind of conduit (different approach in comparison to previous studies [14]) did not change the lack of convergence between experimental and forecasted values of indicator λr. From all values of the length of this kind of conduit only one (69.9 km, in 2007) was selected by the model as the dominant one.…”

“…The main assumption made during the modelling process was that two separate models are built to predict failure rate of distribution pipes and house connections. The approach proposed in this paper is different than that in previous work [14], where one model was created to forecast indicators λ for water mains, distribution pipes and house connections. The results presented in the paper [14] were unsatisfactory from engineering point of view maybe just because one model with three dependent variables was responsible for predicting failure frequency.…”

“…The approach proposed in this paper is different than that in previous work [14], where one model was created to forecast indicators λ for water mains, distribution pipes and house connections. The results presented in the paper [14] were unsatisfactory from engineering point of view maybe just because one model with three dependent variables was responsible for predicting failure frequency. It is the reason why in the following work two models are proposed to check if that changes in the approach may have an influence on the prediction results or maybe MARSplines method is just not suitable for forecasting of failure rate of water pipes.…”

“…This algorithm was recently used in geotechnical and chemical engineering [12,13]. One attempt of using MARSplines method in environmental engineering was done by the author [14] and in the following paper sligthly different approach is proposed to verify hypothesis stated previously [14].…”

A comparative study of statistical and machine learning methods to infer causes of pipe breaks in water supply networks

“…Published studies have predominantly focused on predicting the likelihood of pipe breaks (St. Clair and Sinha 2012;Rajani and Kleiner 2001;Scheidegger, Leitão, and Scholten 2015;Kutyłowska 2018Kutyłowska , 2019. In addition, few studies, have examined the performance of different modelling approaches using the same dataset and over long time periods (Kleiner and Rajani 2012;Yamijala, Guikema, and Brumbelow 2009).…”