Monitoring and control of polymer production line based on machine learning

The work is devoted to the development of an application for monitoring and controlling the state of equipment (extruder) for the petrochemical industry based on sensor readings using a machine learning model. The statistical relationships of the technological process parameters are analyzed, the most significant parameters influencing the occurrence of failures are determined using SHAP values. The hypotheses regarding the effectiveness of various machine learning algorithms in relation to the real problem of predicting the technical state of the extruder are tested. A gradient boosting model has been developed to predict the probability of extruder shutdown due to the formation of polypropylene agglomerates. The developed application allows interpreting the results of the model, which makes it possible to select the most important process parameters that have the greatest impact on the probability of extruder failure, and also proposing a prototype of an extruder monitoring system based on sensor readings using a machine learning model.