Smart Energy Consumption Analysis System for Industrial Machinery Using SHAP-Based Interpretability Models

Abstract

Energy efficiency has become a paramount concern in modern industrial operations, where machinery accounts for a disproportionate share of total facility energy expenditure. This paper presents a Smart Energy Consumption Analysis System (SECAS) designed specifically for industrial machinery environments, leveraging ensemble machine learning techniques combined with SHapley Additive exPlanations (SHAP) to deliver both predictive accuracy and transparent, human-readable interpretability. Industrial machinery datasets — encompassing motor current signatures, vibration telemetry, thermal profiles, and operational load parameters — were collected from three real-world manufacturing plants over a continuous 18-month period. A hybrid model combining Gradient Boosting Machines (GBM) with a Long Short-Term Memory (LSTM) autoencoder was trained to predict near-term energy demand and flag anomalous consumption patterns. SHAP values were then computed to decompose model predictions into feature-level contributions, revealing the dominant influence of spindle speed fluctuations, coolant flow irregularities, and idle-state duration on overall energy waste. The proposed system achieved a Mean Absolute Percentage Error (MAPE) of 3.82% in consumption forecasting and correctly identified 94.7% of energy anomalies with a false-positive rate below 2.1%. Compared to five baseline methods — including classical regression, random forests, and deep convolutional networks — SECAS consistently outperformed across all evaluation metrics. More importantly, the SHAP-driven explainability layer empowers plant operators to act on model recommendations without requiring any machine learning background, a critical feature for real-world adoption. The system was validated through a pilot deployment at a precision engineering facility in Chennai, India, where it contributed to a documented 17.4% reduction in monthly energy costs within a single quarter of operation