Design of Soft sensor based on Machine Learning and its industrial application
Project no. 24039 · Status: Kapanmış (Yayın bekleniyor) · 2022
Summary
<p> Nowadays, the increasing competition in the field of industry focuses on the getting decisions so fast and complete the task as soon as possible. Therefore, in order to prevent human errors, to safe time from the tasks those require human labor and deep laboratory analysis, machines and different types of algorithms are applied and this replacement increases so fast. These developed algorithms can already predict the errors and eliminate them before getting worst cases scenarios related finance and environmental. The soft sensors have capacity to apply created algorithms and these algorithms are created with the help of machine learning. In the today’s world, the data that is used in the machine learning is already collected in all sectors of industry and other fields as well. The importance of data associates with the step of Data Training. More data means much precise results and predictions. When, we look at the usage of soft sensors in the industrial field, the predictions made by soft sensors based on the data collected in the defined time duration can prevent problems those can take the plant into the shut down due to process control related issues. Or much time required laboratorial analysis made and waited to see product purity in order to take some actions to control the system. The content of this thesis is also completely focuses on development and application of soft sensors in the Petrochemistry and Refinery with the help of Python programing language and machine learning. At the end of the thesis, the results obtained from the work will help to predict independent variable more faster and clear. Additionally, the time will be saved from the long lasting laboratory analysis and production of off spec product with saving energy consumption. In the period of Thesis, different type of machine learning methods were applied. As a result, in order to obtain optimized values, the results from different modules were compared and most relevant one was chosen. The models those used are Simple Linear Regression, Multiple linear Regression, Polynomial Regressions, Decision tree regression, Random forest regression and Artificial Neural Network. These algorithms were converted into the project with the help of and Python programing language. The data is collected from the Low Density Polyethylene Production process in order to predict the melting flow index prediction beforehand. There were several possible independent variables collected and then some of them eliminated to avoid wrong training of models. The independent variables were used to train Machine Learning Models and make predictions. All the results are shown at the end of report </p><p><br></p>