Optimizing HVAC systems for semiconductor fabrication: A data-intensive framework for energy efficiency and sustainability

Heating, ventilation, and air conditioning (HVAC) systems play a critical role in maintaining indoor temperature and air quality, directly impacting occupant health, well-being, and productivity. However, their high energy consumption necessitates optimization, for which model predictive control (MPC) has proven highly effective. Grey box modelling has emerged as a valuable tool for assessing indoor air quality and thermal comfort, particularly in real-world scenarios. Traditionally, these models have addressed temperature and air quality separately, which often results in incomplete parameterization and inaccuracies.

The primary objective of this paper is to develop a hybrid grey box model that integrates air and thermal dynamics to improve accuracy in both domains. The methodology involved developing four grey box models to estimate ventilation airflows using indoor CO₂ concentration data and six thermal models to estimate thermal properties and heat gains using indoor temperature data. To ensure accurate parameterization, measurements of outdoor conditions, occupancy, and HVAC operations were incorporated.

The results revealed that models treating infiltration and mechanical ventilation as mutually exclusive (IAQ-3 and IAQ-4) and those integrating ventilation heat gains from estimated airflows (T-6) performed most effectively. This hybrid approach underscores the benefits of incorporating ventilation heat loos or gains, based on airflow estimation derived from indoor air quality (IAQ) models, into thermal modelling, significantly improving accuracy and reducing parameter variability.

The findings demonstrate the potential of this methodology for applications in ventilation management and HVAC optimization. By enhancing energy efficiency and improving indoor air quality, this approach supports the development of healthier, more sustainable indoor environments.

The design, installation, operation, and maintenance of heating, ventilation, and air conditioning (HVAC) systems influence their performance and efficiency. This paper explores HVAC modeling approaches to optimize system efficiency and predict service and replacement lifespans. Optimization accuracy depends on reliable data and the right artificial intelligence (AI) applications to accurately predict outcomes using data-intensive HVAC energy modeling to optimize system performance and efficiency. A Tree-Structured Parzen Estimator-optimized Deep Neural Network (TPE-DNN) is applied to predict temperature (°F) and humidity (%) at the entrance and exit of a make-up air unit (MAU). Random noise, interpolation, and Generative Adversarial Network (GAN) methods are used for data augmentation to enhance model validation. The results demonstrate that the GAN method significantly outperforms traditional data augmentation techniques. For the entrance model, the GAN method achieved an Average Error (AE) of 0.447 (°F), Mean Squared Error (MSE) of 0.313 (°F)², Mean Absolute Error (MAE) of 0.454 (°F), and Mean Absolute Percentage Error (MAPE) of 0.5 %, compared to higher errors with random noise and interpolation. Similarly, for the exit model, the GAN method achieved an AE of 0.447 (°F), MSE of 0.001 (°F)², MAE of 0.024 (°F), and MAPE of 0.03 %, notably lower than those obtained with traditional methods. These results highlight GAN's effectiveness in improving predictive performance, enabling better control of MAU systems' temperature, humidity, air quality, and energy efficiency. The study underscores the potential of advanced AI techniques, particularly GAN, to optimize HVAC operations in dynamic environments like semiconductor manufacturing facilities.

Energy consumption in cooling systems is one of the major environmental burdens in semiconductor manufacturing. Energy-saving measures not only help reduce energy costs but also effectively decrease carbon emissions. These improvements enhance the operational efficiency of the entire supply chain and ultimately benefit downstream enterprises, thereby promoting the sustainable development of the semiconductor supply chain. This study aims to optimize the energy savings in chiller systems in the semiconductor manufacturing. We investigate the interactions between various devices and show how the chiller's operational status affects the temperature setpoint. This study proposes a meta-prediction model to simulate the dynamic behavior of the chiller system, and also employ multi-agent reinforcement learning to support the multi-setpoint control for energy optimization. An empirical study of a semiconductor manufacturer in Taiwan was conducted to validate the proposed model. The results indicate that our developed solution successfully reduced the kilowatts per refrigerated ton (KW/RT) by approximately 2.78% in a practical application.