This dissertation is a study on estimating degradation of high energy steam pipework using modern machine learning techniques. High energy piping systems are very complex to simulate due to the many variables that could in uence the useful life of a component. In this research a hybrid recurrent neural network is created that consists of a combined recurrent neural network and a feed forward neural network. The machine learning model is trained on historical data that has been captured over a six-year time period and is applied to a test dataset to see if any usable patterns exist within the training data. In this research the following variables of the piping system components are used as input to the machine learning model: the operating temperature and pressure time sequence, the distance to the closest anchor point, the distances to neighbouring supports as well as their elevation survey readings and the last known creep damage of the component. The model is created in Python using the Tensor ow library. Two types of recurrent neural networks (RNN) are tested, gated recurrent unit (GRU) and long short term memory (LSTM). The standard gradient descent (GD) algorithm, as well as adaptive gradient descent (ADAGRAD) and adaptive movement estimation (ADAM) are tested. The model was able to predict the classi cation of a component with an accuracy of up to 91% on the training dataset and 56% on the test data set, which is considered to be high given the complexity of the problem. The model is successful in recognising patterns within the data and o ers an automated way to parse large data sets that consist of a temporal and static data mixture. This o ers an approach to make an objective decision on similar complex data driven problems and its application is not constrained to this single problem. The methods applied in this research is expected to perform even better on problems where the frequency of data collection is higher than what is used in this research.
Dissertation (MEng)--University of Pretoria, 2018.