Dataset for assessing HVAC predictive maintenance system performance

Heating, Ventilation and Air Conditioning (HVAC) are crucial installations in the hotel sector. The cost of these facilities often represents a significant percentage of the total building maintenance expenses. One way to reduce these costs is through predictive maintenance. Predictive maintenance aims to keep system components operating optimally and schedule inspections before failures occur. In this paper, we present an Artificial Intelligence-based flexible system for the predictive maintenance of HVAC facilities in hotels. To leverage the advantages of both data-driven models and rule-based models, we propose a model that combines Artificial Neural Networks (ANNs) with a Fuzzy Logic-based expert system. The Fuzzy Logic-based system estimates the probability of upcoming failures in the HVAC based on expert knowledge. The ANN-based system is trained using data generated by the Fuzzy Logic system and then learns adaptively according to the context. This is achieved using a variable number of non-invasive sensors within the HVAC system, providing the Fuzzy Logic (FL) system with the necessary flexibility for accurate operation. Simulations demonstrated strong performance, and the system was successfully tested in a five-star hotel in Seville, Spain.

A total of 10000 samples with corresponding input and output values from the FL system were saved in a spreadsheet. These values were used to create and train the ANNs in MATLAB, establishing the dataset. 6010 samples were generated with standard values of all input variables, varying only one of them randomly. This method creates outputs both without faults and with single faults. An additional 3990 samples were generated using random inputs, allowing for the generation of different possible faults for various input sets. This dataset was divided into three parts: 70% of the samples were used as the training set, 15% as the validation set, and the remaining 15% (675 samples) as the test set

This dataset also contains failure descriptions and the generated code in MATLAB used for simulations

暖通空调（Heating, Ventilation and Air Conditioning，简称HVAC）是酒店行业的关键设施，其运维成本往往占建筑总维护开支的较大比例。降低此类成本的有效途径之一便是预测性维护，该模式旨在保障系统组件处于最优运行状态，并在故障发生前规划检修工作。本文提出了一种基于人工智能（Artificial Intelligence）的柔性系统，用于酒店暖通空调设施的预测性维护。为兼顾数据驱动模型与基于规则模型的双重优势，本文构建了一种将人工神经网络（Artificial Neural Networks，简称ANNs）与基于模糊逻辑的专家系统相结合的混合模型：基于模糊逻辑的子系统可依托专家知识，估算暖通空调系统即将发生的故障概率；基于人工神经网络的子系统则以模糊逻辑系统生成的数据进行训练，并可根据运行场景自适应学习。该系统通过在暖通空调系统内部署可变数量的非侵入式传感器实现上述功能，为模糊逻辑（Fuzzy Logic，简称FL）系统的精准运行提供必要的灵活性。仿真结果表明该系统性能优异，且已在西班牙塞维利亚的一家五星级酒店完成实地测试。

本研究共获取模糊逻辑系统生成的10000组带有对应输入、输出值的样本，并将其存储于电子表格中。利用这些数据在MATLAB平台中构建并训练人工神经网络，最终形成本数据集。其中6010组样本通过固定所有输入变量的标准取值、仅随机改变其中一个变量的方式生成，该方法可同时生成无故障及单故障工况下的输出结果；另有3990组样本通过随机设置输入变量的方式生成，可针对不同输入组合生成多种潜在故障场景。本数据集被划分为三个子集：70%的样本用作训练集，15%用作验证集，剩余15%（共计675组样本）用作测试集。

本数据集还包含故障说明文档以及用于仿真的MATLAB生成代码。