Prediction of machine processing times with machine learning in a factory

Machine learning explores the process of learning from a given data set, uses various techniques based on the desired outcome, and then predicts the target data. Since it enables the prediction of desired information by using different algorithms and calculating the relationship between data, it has become a widely applicable method in various fields and industries with complex data sets. Within the scope of this study, an attempt was made to associate machine learning with an Industry 4.0-based approach used in real-time data collection from the press brake of a company operating in the field of sheet metal, profile, and pipe processing in the automotive and rail systems sector. In line with these data, the time for the twisting process was estimated. By estimating periods close to reality, it was aimed to provide more accurate data input to pricing and efficiency calculation studies than the current situation. In this study, data mining steps were applied to interpret and simplify the data, and twisting operation times were estimated using machine learning prediction algorithms. In the process of selecting the algorithm that makes the most successful prediction, a comparison was made according to the criteria of mean absolute error, 𝑅2, and mean square error, and when the existing data was included in the model, it was seen that the decision tree was the most successful algorithm. It is anticipated that this result may change when new data is added to the model in the future and a different algorithm can make more successful predictions. The study can contribute to the model and data management modules of a decision support system to be used in the pricing phase and can be developed and used in real life in the next stage together with the dialogue management module.