Abstract:

Programmable Logic Controllers (PLC) have been used in almost every field of industry for many years. It is not possible to back up PLC programs in cases such as aging, deterioration or password protection of these PLCs. If the PLC program is deleted as a result of any malfunction or when the PLC needs to be replaced, the program must be rewritten. In such a case, the working steps of the system must be known in detail, and a step that is skipped during the writing of the program can cause major problems in the entire system. In this study, various machine learning algorithms are used to imitate the execution steps of an unknown program running in a PLC. For this purpose, a data log was first created and the input and output information of the PLC was recorded. Then, these input-output data were trained with Machine Learning algorithms. The outputs of these trained algorithms in response to the input data set were monitored in parallel with the PLC outputs. Decision tree, k-nearest neighbor and random forest algorithms were used as machine learning algorithms. Accuracy score was used as the performance measurement metric of the algorithms. At the end of the studies, it was observed that the Random Forest algorithm gave better results than the Machine Learning algorithms.

Keywords: