RII Track-4: Adaptive Fault Detection and Diagnosis Based on Growing Gaussian Mixture Regressions for High-Performance HVAC Systems

在建筑物的整个生命周期中，建筑运营对环境的影响以及资源消耗都很重要。暖通空调（HVAC）系统消耗的能源约占商业建筑总能耗的三分之二。尽管国家努力提高性能和可持续性，但由于设备退化、传感器未校准或控制操作不当，建筑物中的许多现有HVAC系统无法有效运行。这样的问题可能导致高维护成本、居住者不适和浪费能量。建筑物中HVAC系统的故障检测和诊断（FDD）基于测量的系统行为的分析来检测和识别操作故障。FDD技术对于提高建筑物能源效率，减少或消除建筑物因运行故障造成的能源浪费至关重要。当前FDD技术的主要挑战是，可用于创建诊断算法的训练数据不包括测试系统在整个生命周期中经历的所有可能的操作条件。鉴于FDD的训练数据并不覆盖所有的运行条件，建筑HVAC系统的FDD算法必须随着建筑系统和组件的变化而沿着发展。该项目的目标是提高高性能HVAC系统的FDD技术的鲁棒性和效率。 拟议的研究将产生几个更广泛的影响，包括代表性不足的本科生的研究参与，K-12外展活动，以及与其他研究人员共享高性能HVAC系统的实验数据和FDD方法。从这项研究中获得的知识有可能大大提高建筑物的能源效率。 总体研究目标是通过基于自适应机器学习的方法提高故障检测和诊断（FDD）技术的鲁棒性和效率，用于高性能暖通空调（HVAC）系统。这项研究填补了高性能HVAC系统的FDD中的关键知识空白。首先，普渡大学高性能建筑中心高性能暖通空调系统常见故障的实验研究将导致对故障特征的深入了解，包括系统行为以及对能耗和环境条件的影响。虽然对常规HVAC系统的FDD进行了广泛的研究，但对高性能HVAC系统的FDD却很少进行研究。因此，作为本项目的一部分，将获得与高性能HVAC系统中常见故障有关的实验数据，这将成为FDD研究界的宝贵资产。其次，本研究将产生一个自适应FDD方法的基础上增长高斯混合回归高性能的暖通空调系统在商业建筑。传统的FDD方法从在有限的操作条件下测试的训练数据中学习，之后学习停止。这种新的FDD方法适应HVAC运行环境的变化，随着建筑系统和部件的变化而发展，并学会诊断新的故障条件。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Environmental impacts, as well as resource consumption, of building operations are significant throughout the entire life cycle of buildings. Heating ventilation and air conditioning (HVAC) systems consume about two-thirds of the total energy used in commercial buildings. Despite national efforts toward improving performance and sustainability, many existing HVAC systems in buildings do not run efficiently, due to equipment degradation, sensors being out of calibration, or improper control operations. Such problems can result in high maintenance costs, occupant discomfort, and wasted energy. Fault detection and diagnosis (FDD) for HVAC systems in buildings detect and identify operational faults based on the analysis of measured system behaviors. FDD technology is critical to improving building energy efficiency, and reducing or eliminating wasted energy in buildings caused by operational faults. The major challenge in current FDD technology is that the training data available to create diagnostic algorithms do not include all possible operating conditions that the testing systems experience throughout the life cycle. Given that the training data for FDDs does not cover all operating conditions, FDD algorithms for building HVAC systems must evolve along with the changes in building systems and components. The goal of this project is to enhance the robustness and efficiency of FDD technology for high-performance HVAC systems. The proposed research will lead to several broader impacts including research participation of underrepresented undergraduates, K-12 outreach activities, and sharing the experimental data and the FDD method for high-performance HVAC systems with other researchers. The knowledge gained from this research has the potential to significantly enhance building energy efficiency. The overall research goal is to advance robustness and efficiency of Fault detection and diagnosis (FDD) technology through an adaptive machine learning-based approach for high-performance Heating ventilation and air conditioning (HVAC) systems. This research closes critical knowledge gaps in the FDDs for high-performance HVAC systems. First, the experimental study on common faults in high-performance HVAC systems at the Center for High Performance Buildings, Purdue University will result in a thorough understanding of fault features, including system behaviors as well as impacts on energy consumption and environmental conditions. While extensive research has been conducted on the FDD for conventional HVAC systems, the FDD for high-performance HVAC systems has rarely been studied. The experimental data pertaining to common faults in high-performance HVAC systems that will be obtained as a part of this project will, thus, be an invaluable asset to the FDD research community. Second, this research will yield an adaptive FDD method based on growing Gaussian mixture regressions for high-performance HVAC systems in commercial buildings. Traditional FDD methods learn from training data tested under limited operating conditions, after which the learning stops. This new FDD method adapts to the changes in HVAC operating environments, evolves with the changes in building systems and components, and learns to diagnose new faulty conditions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.