Adaptive neural network for dynamic prediction of machine tool energy consumption

Abstract As the “mother machine” of manufacturing, machine tools are extensively utilized, but suffer from high energy consumption and low energy efficiency. Energy consumption prediction for machine tools holds significant importance for energy planning in the manufacturing industry. Some scholars have achieved the prediction of machine tool energy consumption using data-driven modeling methods. However, these studies are based on single processing conditions, requiring a substantial amount of data support to achieve accurate predictions under specific conditions. In practical processing, machine tools have multiple working conditions, making it impractical to obtain prediction models for each condition. To this end, this paper proposes an adaptive neural network algorithm for energy consumption prediction under different working conditions. When machine tool conditions change, the existing prediction model can be adjusted by acquiring a limited quantity of current condition data to obtain a prediction model for the current condition. It reduced data collection time and lowered model reconstruction costs. A milling experiment was conducted to validate the practicality and feasibility of the proposed algorithm. The unadapted neural networks and other algorithmic models exhibit error rates of over 15% on the training set, but the adaptive neural networks maintain an error rate of essentially below 8%.