A diagnosis and classification method for energy efficiency anomalies in mechatronic equipment based on multi-condition inherent energy efficiency benchmark

Ensuring the high-efficiency operation of mechatronic equipment has become a key pathway for manufacturing enterprises to reduce costs and achieve green transformation. However, the coupled effects of internal self-degradation and external human operation make energy efficiency anomalies prone to misjudgment, thereby exacerbating energy consumption and economic losses. The aim is to diagnose and classify energy efficiency anomalies in mechatronic equipment more accurately under varying working conditions. To this end, a novel diagnosis and classification method for energy efficiency anomalies is proposed. Firstly, composite filtering and state parameter estimation are employed to preprocess the raw data. Then, a Neural Supervised Conditional Inference Tree (NSCIT) algorithm is proposed to establish multi-condition inherent energy efficiency benchmark by integrating maximum energy efficiency potential. Second, a matching function is constructed using a Negative Selection Algorithm (NSA) to diagnose real-time energy efficiency anomalies. Furthermore, by integrating a Convolutional Neural Network (CNN) with a Bidirectional Long Short-Term Memory Network (BiLSTM), the time series features of deviations between real-time energy efficiency and the inherent efficiency benchmark are extracted and analyzed. Through a fully connected layer, a mapping relationship between the time series features vector and the type of working condition anomaly is established, achieving the identification of energy efficiency anomaly types under multi-conditions. Finally, the method is validated in CNC milling experiments to accurately identify energy efficiency anomaly types under multi-conditions, supporting high-quality and specialized energy efficiency operation and maintenance decision-making for mechatronic equipment.