Artificial data set for benchmarking pre-processing algorithms for distributed fiber optic strain data

Distributed strains sensing (DSS) with distributed fiber optic sensors (DFOS) has great potential for structural health monitoring (SHM). Raw DSS data might contain different types of disturbances caused by the measurement principle of DFOS. The disturbance types are (i) misreadings called strain reading anomolies (SRA), (ii) missing values called dropouts, and (iii) noise. Hence, pre-processing (the process of removing or reducing the disturbances) is key for a reliable evaluation of DSS data. Many different pre-processing approaches/algorithms exist. The assessment, how well an algorithms performs in removing the disturbances is done by benchmarking. This judgement requires a known "ground truth" (disturbance free signal). As all measurements show noise, this benchmarking needs to be carried out on an artifical data set. The aim of this benchmark data set is to simulate realistic DSS data. The characteristics of the benchmark data set is described in in detail in the accompanying paper available at [10.3390/s24237454](https://doi.org/10.3390/s24237454). To simulate different use cases, the data set contains five scenarios. SRAs, dropouts and noise are simulated using simple random processes. The values for SRAs are extracted from the data set available at [10.25532/OPARA-671](https://doi.org/10.25532/OPARA-671). This dataset is available at [10.25532/OPARA-644](https://doi.org/10.25532/OPARA-644) and accompanies the paper [10.3390/s24237454](https://doi.org/10.3390/s24237454).

分布式应变传感（Distributed Strains Sensing, DSS）依托分布式光纤传感器（Distributed Fiber Optic Sensors, DFOS），在结构健康监测（Structural Health Monitoring, SHM）领域具备巨大应用潜力。原始DSS数据通常会包含由DFOS测量原理引入的多种干扰类型，具体包括：（i）被称为应变读数异常（Strain Reading Anomalies, SRA）的误读数据，（ii）被称为数据丢包（Dropouts）的缺失值，以及（iii）噪声。因此，预处理——即去除或削弱上述干扰的过程——是实现可靠DSS数据评估的核心关键。目前已存在多种不同的预处理方法与算法，而评估某一算法对干扰的去除效果需通过基准测试完成，该评判过程需要已知的“真值（ground truth）”，即无干扰信号。由于所有实测数据均存在噪声，此类基准测试需基于人工数据集开展。本基准数据集旨在模拟真实场景下的DSS数据，其详细特性已在随附的论文[10.3390/s24237454](https://doi.org/10.3390/s24237454)中进行了全面阐述。为模拟不同的应用场景，本数据集包含五类工况。应变读数异常、数据丢包与噪声均通过简单随机过程进行模拟，其中应变读数异常的取值源自公开数据集[10.25532/OPARA-671](https://doi.org/10.25532/OPARA-671)。本数据集可通过[10.25532/OPARA-644](https://doi.org/10.25532/OPARA-644)获取，并与论文[10.3390/s24237454](https://doi.org/10.3390/s24237454)配套发布。