CHANGEPOINT DETECTION WITH PYTHON ON SENSOR DATA IN WELDING PROCESSES

In a machining process, there are several stages of operation involved. The sensors are used to collect the data in all the stages at the desired intervals of time of the user. Once there is a large amount of data set, finding the stages becomes complex. The stages can be identified by the Changepoint, but the identification of Changepoint is a challenge. Changepoint detection is the estimation of breakpoints at which the statistical properties of the observation change. This Changepoint gives the inference that the stage has started/ended. Pruned Exact Linear Time (PELT) is the method adopted for detecting changepoints, which is through minimizing the cost function over possible numbers and locations of changepoints. This method performs efficiently compared to the other existing methods. To understand it better in the simulation study a comparison of PELT and dynamic programming is also made to validate the efficiency. This thesis aims to present a framework to estimate the Changepoints and segment the process using statistical properties of the observations made. The first part of the framework is identify the changepoints and segment the processes, while the second part deals with the development of different types of algorithms and compared to find the best result. The operation is carried out in Colabatory, an IDE for Python released by Google to enable machine-learning with storage on the cloud. Ruptures, a python library for offline change point detection has been used, its algorithm exact and approximate detection for various parametric and non-parametric models. Ruptures provide methods for the analysis and segmentation of non-stationary signals. To validate the algorithm, the raw data set of a milling machine has been used. After the validation, it is aimed to be incorporated in the Machine monitoring of the Welding machine.