Jino Mathew

Anomaly detection in time series data using a combination of wavelets, neural networks and Hilbert transform

In this paper, a new signal processing algorithm for detecting anomalies in time series data is proposed. Real time detection of anomalies is crucial in structural health monitoring applications as it can be used for an early detection of structural damage as well as for discovery of abnormal operating conditions that can shorten a structures life. A new algorithm - a combination of wavelets, neural networks and Hilbert transform - is presented and discussed in this study. The algorithm has been evaluated for a number of benchmark tests, commonly used in the literature, and has been found to perform robustly.

Optimised Neural Network Prediction of Residual Stress Profiles for Structural Integrity Assessment of Pipe Girth Welds

Defect assessment procedures such as BS7910, R6, FITNET and API 579-1 provide simplified estimates or upper bound profiles that can be used to characterize residual stresses present in a weld. Some of the bounding through-thickness profiles used in these procedures are designed based on expert judgment and examination of residual stress measurements that exhibit wide scatter. As a consequence, structural integrity assessment of defects in welded components can be overly conservative by a large margin, and may lead to unnecessary and costly repair or inspection. This paper presents a novel approach based on artificial neural networks to predict residual stress profiles in austenitic stainless steel pipe girth welds. The network is trained using a set of baseline experimental residual stress data and then validated using previously unseen data. The committee of networks has been optimized using a Bayesian approach and the upper bound curve is determined from the histogram network of output distributions. The performance and suitability of the neural network approach is discussed by comparison with stress profiles recommended in the R6 procedure and followed by an assessment of whether the use of neural network bounding profiles can lead to non-conservative estimates of stress intensity factor in fracture assessments.Copyright

Probabilistic kernel machines for predictive monitoring of weld residual stress in energy systems

Abstract Predictive monitoring supports the a priori scheduling of critical component maintenance and contributes significantly in attaining a safe yet economic operation and management of complex energy systems by mitigating the risk of accidents and minimizing the number of operational pauses. The current work studies the learning ability of probabilistic kernel machines, and more particularly of Gaussian Processes (GP) equipped with various kernels for the estimation of weld residual stress profiles of stainless steel pipe welds. The GP models are tested on experimentally-obtained data of axial and hoop residual stresses in two different stainless-steel pipes. The results exhibit the ability of GP to accurately predict the weld residual stress profile in the axial and hoop direction by providing a predictive distribution, i.e., mean and variance values. Furthermore, performance of GP is compared to a non-probabilistic kernel machine, such as support vector regression (SVR) equipped with the same kernels, and to multivariate linear regression (MLR). Comparison results exhibit the robustness of GP over SVR and MLR with respect to prediction accuracy of weld residual stress in terms of root mean square error. With respect to a second metric, namely, correlation coefficient between measured and predicted values, GP is superior to SVR and MLR in the majority of the cases.

Modelling and Measuring Residual Stresses in Pipe Girth Welds: Lessons From the Style Framework 7 Project

A number of girth-welded pipe mock-ups have been manufactured and investigated during the STYLE project, using a wide range of measurement techniques accompanied by extensive finite element simulation campaigns. This paper gives an overview of the work carried out and presents preliminary conclusions on the performance of finite element weld residual stress simulation techniques in the different mock-up designs.Copyright

Prediction of welding residual stresses using machine learning: Comparison between neural networks and neuro-fuzzy systems

Abstract Safe and reliable operation of power plants invariably relies on the structural integrity assessments of pressure vessels and piping systems. Welded joints are a potential source of failure, because of the combination of the variation in mechanical properties and the residual stresses associated with the thermomechanical cycles experienced by the material during welding. This paper presents comparative studies between methods based on artificial neural networks (ANN) and fuzzy neural networks (FNN) for predicting residual stresses induced by welding. The performance of neural network and neuro-fuzzy systems are compared based on statistical indicators, scatter plots and several case studies. Results show that the neuro-fuzzy systems optimised using a hybrid technique can perform slightly better than a neural network trained using Levenberg-Marquardt algorithm, primarily because of the inability of the ANN approach to provide conservative estimates of residual stress profiles. Specifically, the prediction accuracy of the neuro-fuzzy systems trained using the hybrid technique is better for the axial residual stress component, with root mean square error (RMSE), absolute fraction of variance (R2) and mean absolute percentage error (MAPE) error of 0.1264, 0.9102 and 22.9442 respectively using the test data. Furthermore, this study demonstrates the potential benefits of implementing neuro-fuzzy systems in predicting residual stresses for use in structural integrity assessment of power plant components.

Bounding Residual Stress Intensity Factor Profiles for Fracture Assessment of Pipe Girth Welds

The effect of residual stress on potential crack growth and fracture in welded structures is usually assessed through its contribution to the stress intensity factor (SIF) for the crack size and shape of interest. The idea of defining bounding residual SIF profiles for surface breaking circumferential cracks in pipe butt welds was presented at ASME PVP2013. The limiting profiles were based on through-thickness residual stress measurements for eight pipe girth welds. This paper presents new axial residual stress measurements made using the contour method for an Esshete 1250 stainless steel pipe girth weld. A wide variation in the through-wall distribution of axial residual stress around the circumference of the pipe is observed which has a significant effect on calculated values of SIF for postulated surface breaking circumferential cracks. Nonetheless, SIFs based on all of the new measurements (a total of 14 profiles) are comfortably bounded by the simple SIF prescriptions previously published.Copyright