This paper presents a performance evaluation of a 140 kW solar array installed on the rooftop of the Mountain Line Transit Authority (MLTA) building in Morgantown, West Virginia (WV), USA, covering the period from 2013 to 2024. The grid-connected photovoltaic (PV) system consists of 572 polycrystalline PV modules, each rated at 245 watts. The study examines key performance parameters, including annual electricity production, average daily and annual capacity utilization hours (CUH), current array efficiency, and performance degradation. Monthly ambient temperature and global tilted irradiance (GTI) data were obtained from the NASA POWER website. During the assessment, observations were made regarding the tilt angles of the panels and corrosion of metal parts. From 2013 to 2024, the total electricity production was 1588 MWh, with an average annual output of 132 MWh. Over this 12-year period, the CO₂ emissions reduction attributed to the solar array is estimated at 1,413,497 kg, or approximately 117,791 kg/year, compared to emissions from coal-fired power plants in WV. The average daily CUH was found to be 2.93 h, while the current PV array efficiency in April 2024 was 10.70%, with a maximum efficiency of 14.30% observed at 2:00 PM. Additionally, an analysis of annual average performance degradation indicated a 2.28% decline from 2013 to 2016, followed by a much lower degradation of 0.17% from 2017 to 2023, as electricity production data were unavailable for most summer months of 2024.

Industrial energy assessment programs generate large volumes of facility-level recommendations, yet the engineering knowledge embedded in these reports often remains stored as unstructured narrative text and is difficult to reuse for data-driven decision support. This study investigates whether machine-learning and NLP models can systematically understand, differentiate, and retrieve knowledge from industrial assessment scenarios. Because no suitable benchmark existed for this purpose, we constructed a labeled dataset of 812 unique industrial energy-efficiency scenarios extracted from 294 WVU Industrial Training and Assessment Center assessments conducted between 2010 and 2025. Each scenario is annotated with Energy Conservation Measure category, savings type, efficiency status, savings magnitude, and resolution time. Using this dataset, we define five classification benchmarks and one case-based retrieval benchmark to evaluate industrial audit language understanding. Baseline experiments compare TF-IDF with Logistic Regression and fine-tuned RoBERTa-base. Results show that industrial assessment text contains structured, machine-learnable information: RoBERTa-base achieved the strongest Macro-F1 for ECM category classification (0.809), binary efficiency-status classification (0.984), and energy-saving magnitude classification, while TF-IDF with Logistic Regression remained highly competitive and achieved the strongest retrieval performance, reaching Recall@5 of 0.969 under ECM-category relevance. Overall, this work shows that AI models can extract actionable structure from industrial audit text and provides a reproducible foundation for classification, recommendation retrieval, knowledge reuse, and decision support in industrial energy efficiency.

Industrial energy assessment programs generate large volumes of facility-level recommendations, yet the engineering knowledge embedded in these reports often remains stored as unstructured narrative text and is difficult to reuse for data-driven decision support. This study investigates whether machine-learning and NLP models can systematically understand, differentiate, and retrieve knowledge from industrial assessment scenarios. Because no suitable benchmark existed for this purpose, we constructed a labeled dataset of 812 unique industrial energy-efficiency scenarios extracted from 294 WVU Industrial Training and Assessment Center assessments conducted between 2010 and 2025. Each scenario is annotated with Energy Conservation Measure category, savings type, efficiency status, savings magnitude, and resolution time. Using this dataset, we define five classification benchmarks and one case-based retrieval benchmark to evaluate industrial audit language understanding. Baseline experiments compare TF-IDF with Logistic Regression and fine-tuned RoBERTa-base. Results show that industrial assessment text contains structured, machine-learnable information: RoBERTa-base achieved the strongest Macro-F1 for ECM category classification (0.809), binary efficiency-status classification (0.984), and energy-saving magnitude classification, while TF-IDF with Logistic Regression remained highly competitive and achieved the strongest retrieval performance, reaching Recall@5 of 0.969 under ECM-category relevance. Overall, this work shows that AI models can extract actionable structure from industrial audit text and provides a reproducible foundation for classification, recommendation retrieval, knowledge reuse, and decision support in industrial energy efficiency.

Industrial facilities are under increasing pressure to reduce energy consumption, lower greenhouse gas emissions, and demonstrate measurable progress toward sustainability and environmental-performance goals. Boiler and steam systems are major fuel consuming assets in many facilities, yet facility managers often lack accessible tools for comparing the energy, cost, and emissions impacts of alternative boiler-efficiency measures before implementation. This study develops an Excel Visual Basic for Applications (VBA) based temporal simulation tool for evaluating boiler energy efficiency projects under facility specific operating conditions. The model estimates baseline boiler energy use and CO2 emissions from user defined operating parameters, applies selected efficiency measures from specified implementation dates, and calculates daily and annual changes in fuel consumption, energy cost, and carbon emissions. Nine scenarios were evaluated, including eight individual boiler-efficiency measures and one combined scenario. Under the selected baseline operating conditions, condensate recovery produced the largest energy reduction among the individual measures, followed by burner replacement, air-fuel ratio optimization, steam pressure reduction, blowdown heat recovery, blowdown rate reduction, flue gas heat recovery, and insulation of exposed hot surfaces. The combined scenario, which incorporated condensate recovery, air-fuel ratio optimization, and steam pressure reduction at different implementation dates, produced the highest overall savings. The model was verified using 12 months of natural gas utility bill data from a manufacturing facility. The simulated annual energy consumption differed from the utility bill total by 1.48%, while monthly absolute percentage errors ranged from 1.48% to 12.09%. These differences were attributed to production variation, weather effects, simplified load factor assumptions, and real world operating variability. The proposed tool provides a transparent and low cost decision support framework for preliminary boiler project evaluation, energy savings estimation, emissions accounting, and ESG oriented environmental performance tracking.