Autonomous underwater vehicle fault diagnosis dataset

Daxiong Ji, Xin Yao, Shuo Li, Yuangui Tang, Yu TianData 2021 · DOI

TL;DR

This paper introduces a new dataset for autonomous underwater vehicle (AUV) fault diagnosis, containing 1225 samples across five fault types. This dataset aims to provide a standardized benchmark for developing and validating model-free fault diagnosis methods for AUVs.

What Problem Does This Paper Address?

Diagnosing faults in autonomous underwater vehicles (AUVs) is crucial for their reliable operation, especially in complex and remote underwater environments. However, developing and rigorously testing fault diagnosis systems requires comprehensive, real-world data. Existing datasets may be limited in scope, not representative of actual AUV operations, or lack standardized formats for benchmarking. This paper addresses this gap by creating and releasing a dedicated dataset for AUV fault diagnosis, enabling researchers to develop and compare model-free diagnostic approaches.

Methods

The authors collected data from a small quadrotor AUV named 'Haizhe'. They simulated five distinct fault types: normal operation, slight propeller damage, severe propeller damage, depth sensor failure, and increased load. For each fault type, 'Haizhe' was operated underwater for 10-20 seconds, collecting state data. This state data, along with the corresponding fault type as the label, forms the dataset. The data was split into training and testing sets using random stratified sampling, with the test set comprising 20% of the total data. The dataset was specifically designed to validate a model-free fault diagnosis method previously proposed by the authors.

Datasets

The primary contribution of this paper is the introduction of a new dataset for AUV fault diagnosis.

  • Dataset Name: Autonomous underwater vehicle fault diagnosis dataset.
  • Size: 1225 data samples.
  • Fault Types: 5 (normal state, slight propeller damage, severe propeller damage, depth sensor failure, load increase).
  • Data Format: CSV files containing state data.
  • Structure: The dataset is divided into "train" and "test" folders, each containing subfolders for the five fault types.
  • Relation to Existing Benchmarks: This dataset is presented as a new benchmark specifically for AUV fault diagnosis, aiming to provide a standardized resource for researchers in this domain.

Key Insights and Ideas

  • The Importance of Real-World Data: The paper highlights the necessity of collecting data from actual AUV operations to accurately train and validate fault diagnosis systems, rather than relying solely on simulations.
  • Model-Free Diagnosis Potential: By releasing this dataset, the authors aim to foster research into model-free fault diagnosis methods, which can be more robust to uncertainties and complexities in AUV dynamics.
  • Standardization for Benchmarking: The creation of a structured dataset with clear labels and a defined split allows for more objective comparison of different fault diagnosis algorithms.

Key Results

The paper's main contribution is the dataset itself, rather than specific quantitative results from its application. The dataset contains 1225 samples, divided into training and testing sets (80% train, 20% test). The data includes 17 columns of state information, such as time, motor PWM signals, depth, pressure, battery voltage, roll, pitch, yaw angles, accelerations (ax, ay, az), and angular velocities (w_row, w_pitch, w_yaw). The dataset was used to validate a model-free fault diagnosis method proposed in a prior publication by the authors.

Limitations and Future Work

  • Dataset Size: The authors acknowledge that the submitted dataset is "not big enough" and state their intention to "gradually collect more samples."
  • Single Fault Type: The experiments were conducted with only one fault type present at a time, meaning the dataset does not include scenarios with concurrent multiple faults.
  • Simulated Depth Sensor Fault: The depth sensor fault was artificially introduced by adding a bias to the sensor readings, rather than being a hardware failure. This might affect the direct applicability to diagnosing actual hardware malfunctions.
  • Future Work: The authors hope that more researchers will use this dataset to propose better diagnosis models and will continue to collect more samples to expand the dataset. They also suggest that researchers can use statistical knowledge or machine learning for data mining on this dataset.

Summary updated

Abstract

The dataset contains 1225 data samples for 5 <i>fault types</i> (labels). We divided the dataset into the training set and the test set through random stratified sampling. The test set accounted for <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mrow><mml:mn>20</mml:mn> <mml:mo>%</mml:mo></mml:mrow> </mml:math> of the total dataset. Our experimental subject is 'Haizhe', which is a small quadrotor AUV developed in the laboratory. For each <i>fault type</i>, 'Haizhe' was tested several times. For each time, 'Haizhe' ran the same program and sailed underwater for 10-20 s to ensure that <i>state data</i> was long enough. The <i>state data</i> recorded in each test were then used as a data sample, and the corresponding <i>fault type</i> was the true label of the data sample. The dataset was used to validate a model-free fault diagnosis method proposed in our paper [1] and the complete dynamic model of 'Haizhe' AUV was reported in [2].
BibTeX
@inproceedings{ji2021autonomous,
  author = {Daxiong Ji and Xin Yao and Shuo Li and Yuangui Tang and Yu Tian},
  title = {{Autonomous underwater vehicle fault diagnosis dataset}},
  booktitle = {Data},
  year = {2021},
  doi = {10.1016/j.dib.2021.107477},
}
Figure