Создан нейросетевой алгоритм, который улучшает качество токарной обработки.

The article has been published in the Journal of Digital Science. The research was conducted as part of the strategic academic leadership program "Priority 2030".

When processing metal blanks on standard CNC machines, the program is adjusted to specific parameters. Factors such as the hardness of the material being processed, the thickness of the layer to be removed, and many other indicators that affect the quality of the final product are taken into account. However, during the metal cutting process, uncontrolled random changes in the properties of cutting tools occur. Additionally, each subsequent blank from the processed batch exhibits differences in surface structure and hardness. All of this necessitates constant monitoring by the machine operator.

Adaptive control, unlike conventional systems, provides automatic adjustment of process parameters to changing conditions.

“Based on the information received about the current state of the processing operation, the system automatically increases or decreases the amount of metal removed from the blank, thereby maintaining the threshold value of a given parameter, such as cutting force. In more complex cases, it ensures the achievement of optimal values for accuracy, productivity, or cost-effectiveness of processing the blanks,” explains Vladimir Oniskiv, Associate Professor at the Department of Computational Mathematics, Mechanics, and Biomechanics at PNIPU, Candidate of Technical Sciences.

Artificial intelligence methods are increasingly being applied in adaptive control systems for turning processes. However, the impact of tool wear on the roughness of the processed surface is still insufficiently studied. Researchers at Perm Polytechnic have developed an algorithm utilizing a neural network that ensures the necessary level of roughness while enhancing cutting productivity.

“We hypothesized that this indicator depends on the degree of wear of the cutting tool, which is determined by the current level of vibration. Our trained neural network predicts the roughness value based on the energy levels of signals from vibration sensors at specific frequencies, given the parameters of the cutting mode. Based on this, we developed an algorithm that, upon receiving a signal indicating that the maximum allowable roughness level has been reached, will adjust the tool feed parameters to appropriate levels,” explains Vladimir Oniskiv.

The polytechnic researchers emphasize that the algorithm meets the conditions for optimal control, as the processing begins with the highest feed rate and gradually decreases. The system ensures the maximum possible amount of metal removal at a specified surface roughness, significantly enhancing the productivity of metal processing.

The algorithm proposed by PNIPU researchers has already been tested on real data, confirming its applicability in the development of an intelligent information system for adaptive control of the turning process.