Remaining useful life prediction and bug triaging using machine learning techniques

Abstract

In the field of engineering, it’s become important to understand machines and components of it, not only in terms of how they are performing currently but also how their performance degrades over time. Breakdown of a machine has a huge impact on the industrial system and can lead to tragic situation if regular maintenance is not carried out [8]. Predictive maintenance observes machines behaviour at periodic intervals or continuously and based on this data predicts Remaining Useful Life (RUL) of machines, which enables maintenance before breakdown of machines. In the past, a lot of research has been performed to compare the performance of various RUL prediction techniques such as Kalman Filter [9], Multi-Layer Perceptron (MLP), Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) [10] and even adversarial approaches also. In this thesis, We have proposed to use Attention along with Long Short-Term Memory (LSTM) [11]based approach to improve the RUL prediction. Bug tracking tools such as JIRA, Bugzilla receives hundreds of bugs daily. Each of this bugs needs to be triaged means assign it to developer or team. This triaging process consumes a significant amount of time and resources. This time spent of triaging can be significantly invested in improvement of software. There comes the need of automatic bug triaging. In the past, different information like as title, description and developer’s work history [12][13] are used to learn the features like tf-idf, tf, Bag of Word (BOW) and keywords.[14] This features are used to train various classifiers such as Support Vector Machine (SVM) and Naive Bayes.[15] We have proposed two approaches for Bug Triaging. One is feature learning using Attention based Deep Bidirectional Long Short-Term Memory (A-DBLSTM) and classification using softmax classifier. Second approach is based on transfer learning.