Vibrations of Injection Machine

Production quality control is an issue of great importance in the industry. Generating defective products leads to wasted time and money. For this reason, we have attempted to develop a production control system using computational artificial intelligence methods. The system, in its current version, has been developed and tested, using the example of controlling the operation of an injection moulding machine producing plastic elements. The diagnostic system consists of sensors collecting vibration signals of the machine, a data acquisition module, a data transfer module, and a decision-making module. We obtain descriptors from time series data obtained from sensors using established methods of signal processing. Then, two feature selection algorithms, Boruta and MDFS, were used to identify relevant descriptors. Several machine learning algorithms were used to predict the probability of malfunction of the production process. Finally, the prediction results for a series of several last production cycles were aggregated to decide about raising the alarm. Fifteen descriptors that were related to the quality of the process have been identified. Random Forest machine learning algorithm proved to be the most effective; however, its predictions didn't allow us to make a decision based on a single production cycle. The decision strategy is based on the aggregation of predictions for a series of cycles, which leads to the sensitivity and selectivity appropriate for practical use.