An Interpretable Production-Aware Rule-Based Decision-Support System for Intelligent Energy Management in Smart Factories

Abstract

Smart factories need transparent methods that distinguish high energy demand caused by production from poor energy performance. This study presents an interpretable decision-support framework based on daily total energy consumption and product-referenced specific energy consumption (SEC) for three distinct anonymized products. The analysis used 436 daily records collected between 2019 and 2026 from a single manufacturing plant. After the data-quality audit, 401 valid energy records were retained. Normality was rejected in seven of the eight annual datasets. Chronological Q90 and Q95 thresholds were therefore used as the main approach and compared with limits based on the mean and standard deviation and on the median and MAD. A complete rule matrix assigned four recommendation levels, while a separate robust score identified unusual operating days. Across 907 evaluations, activation rates ranged from 16.96% to 26.95%, Level 3 recommendations from 1.77% to 2.60%, and anomaly rates from 6.65% to 8.48%. Confidence intervals and sensitivity analyses showed that the main conclusions remained stable across the analytical choices. The framework supports monitoring, inspection, scheduling, and prioritized operator review.