ArticlePDF Available

Autonomous Underwater Vehicle Fault Diagnosis Dataset

Authors:
Daxiong Ji at Zhejiang University
Daxiong Ji
Xin Yao at Zhejiang University
Xin Yao
Shuo Li at Jilin University
Shuo Li
Yuangui Tang
Yuangui Tang
Yuangui Tang
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Show all 5 authorsHide
Download full-text PDF
Read full-text
Download citation

Abstract and Figures

The dataset contains 1225 data samples for 5 fault types (labels). We divided the dataset into the training set and the test set through random stratified sampling. The test set accounted for 20% of the total dataset. Our experimental subject is ‘Haizhe’, which is a small quadrotor AUV developed in the laboratory. For each fault type, ‘Haizhe’ was tested several times. For each time, ‘Haizhe’ ran the same program and sailed underwater for 10-20 seconds to ensure that state data was long enough. The state data recorded in each test were then used as a data sample, and the corresponding fault type 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].
Figures - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Available via license: CC BY 4.0
Content may be subject to copyright.
Data in Brief 39 (2021) 107477
Contents lists available at ScienceDirect
Data in Brief
journal homepage: www.elsevier.com/locate/dib
Data Article
Autonomous underwater vehicle fault
diagnosis dataset
Daxiong Ji
a , b , ∗
, Xin Yao
a , b
, Shuo Li
c
, Yuangui Tang
c
, Yu Tian
c
a
The Key Laboratory of Ocean Observation-Imaging Test bed of Zhejiang Province, The Institute of Marine Electronic
and Intelligent System, Ocean College, Zhejiang University, China
b
The Engineering Research Center of Oceanic Sensing Technology and Equipment, Ministry of Education, Zhoushan
3160 0 0, China
c
Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang, Liaoning 110 0 16 , China
a r t i c l e i n f o
Article history:
Received 8 April 2021
Revised 8 September 2021
Accepted 9 September 2021
Available online 14 October 2021
Keywo rds:
Fault diagnosis
Autonomous underwater vehicles (AUV)
Model-free
State data
Fault type
a b s t r a c t
The dataset contains 1225 data samples for 5 fault types (la-
bels). We divided the dataset into the training set and the
test set through random stratified sampling. The test set ac-
counted for 20% of the total dataset. Our experimental sub-
ject is ‘Haizhe’, which is a small quadrotor AUV developed in
the laboratory. For each fault type , ‘Haizhe’ was tested sev-
eral times. For each time, ‘Haizhe’ ran the same program and
sailed underwater for 10–20 s to ensure that state data was
long enough. The state data recorded in each test were then
used as a data sample, and the corresponding fault type 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] .
©2021 The Author(s). Published by Elsevier Inc.
This is an open access article under the CC BY license
( http://creativecommons.org/licenses/by/4.0/ )
∗Corresponding author at:The Key Laboratory of Ocean Observation-Imaging Testb ed of Zhejiang Province, The Insti-
tute of Marine Electronic and Intelligent System, Ocean College, Zhejiang University, China.
E-mail addresses: jidaxiong@zju.edu.cn (D. Ji), xyao@zju.edu.cn (X. Yao ), shuoli@sia.cn (S. Li), tyg@sia.cn (Y. Tang) ,
tiany@sia.cn (Y. Tian).
https://doi.org/10.1016/j.dib.2021.107477
2352-3409/© 2021 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license
( http://creativecommons.org/licenses/by/4.0/ )
2 D. Ji, X. Yao and S. Li et al. / Data in Brief 39 (2021) 107477
Specifications Tabl e
Subject Ocean and Maritime Engineering
Specific subject area Fault Diagnosis of Autonomous Underwater Vehicles
Type of data Tabl e
How data were acquired We used ‘Haizhe’ AUV as the experimental subject. ‘Haizhe’ is installed with
4 brushless motors (SUNNYSKY A2212 KV980 II), 4 propellers (Three-bladed
Propeller Outer Diameter: 55mm,Thread Pitch: 80mm), 4 electronic speed
control (HOBBYWING Skywalker 20A),
1 depth sensor (MS5803-01 BA) , 1 nine-axis inertial measurement unit
(GY-MPU9250), and 1 microcontroller
unit (STM32F407VET6).
Data format Raw
Parameters for data collection We set five common fault types for ‘Haizhe’, including normal state, slight
damage to the propeller, severe damage to the propeller,
failure of the depth
sensor, and load increase. In the experiment, ‘Haizhe’ had only one fault type
at the same time, and there was no multi-failure concurrency. The fault of
depth sensor was not the hardware damage but artificially added a bias item
when reading the pressure va lue, which made the calculated depth value
deviate from the true value. For example, when the true depth was 0.5m, the
depth calculated by the sensor wa s 0.6m. ‘Haizhe’ sailed underwater for 10-20
seconds each time.
Description of data collection Firstly, the fault type was set and recorded for the ‘Haizhe’. Then, the initializer
was executed to check whether each component could work properly or not.
After that, ‘Haizhe’ executed the main program and turned on the function of
data recording. And then ‘Haizhe’ began sailing underwater. It is worth noting
that different behavioral responses would be generated under the influence of
different fault types , but the main program did not change. After completing
the main program, ‘Haizhe’ would stop data recording and automatical ly ro se
to the surface. Finally, the file system of ‘Haizhe’ would save the state data as a
text file (also called a data sample). For each fault type, ‘Haizhe’ was
tested
several times. The state data recorde d in each test were then used as a data
sample, and the corresponding fault type was the true label of the data sample.
Data source location Institution: The Institute of Marine Electronics and Intelligent Systems, Ocean
College, Zhejiang University City/Town/Region: Zhoushan Country: China
Data accessibility https://doi.org/10.17632/7rp2pmr6mx.1
Related research article https://doi.org/10.1016/j.oceaneng.2021.108874 .
Value of the Data
• Our paper proposed a diagnosis model for AUV, which could learn the potential pattern be-
tween state data and fault type from the dataset. We hope that more researchers will pay
attention to our approach and propose a better diagnosis model. The submitted dataset is
recorded in the experiment, which can be used as a standard dataset to verify the perfor-
mance of the diagnosis model. Although, the submitted dataset is not big enough, we will
gradually collect more samples.
• Those who research on fault diagnosis of autonomous underwater vehicle or want to analyze
the correlation between state data and fault type can benefit from this dataset.
• The submitted dataset includes training set and test set. Researchers can train their diagno-
sis model from training set and use the test set to validate the performance of the trained
model. In addition, researchers can also use statistical knowledge or machine learning for
data mining on this dataset.
Data Description
In the submitted dataset, the folder (named “Dataset”) contains two folders (one is named
“train”, another is named “test”). “train” means training dataset, while “test” means test dataset.
D. Ji, X. Yao and S. Li et al. / Data in Brief 39 (2021) 107477 3
Both folders contain 5 folders: “AddWeight”, “Normal”, “PressureGain constant”, “PropellerDam-
age bad” and “PropellerDamage slight”. “AddWeight” corresponds to load increase ( fault type );
“Normal” corresponds to normal state; “PressureGain constant” corresponds to failure of the
depth sensor; “PropellerDamage bad” corresponds to severe damage to the propeller; “Propeller-
Damage slight” corresponds to slight damage to the propeller.
Each fault type folder contains data samples. And each sample is a.csv file, which records
state data of ‘Haizhe’ over a certain period of time. The name of fault type folder represents the
true label of the sample. There are 17 columns in a.csv file. The name and description of each
column is listed as below:
• time: The absolute time for ‘Haizhe’ to record data.
• pwm1: Duration (in microseconds) of high level in 10 0 Hz PWM wave. It is the control signal
used to control the Motor 1.
• pwm2: Duration (in microseconds) of high level in 10 0 Hz PWM wave. It is the control signal
used to control the Motor 2.
• pwm3: Duration (in microseconds) of high level in 10 0 Hz PWM wave. It is the control signal
used to control the Motor 3.
• pwm4: Duration (in microseconds) of high level in 100Hz PWM wave. It is the control signal
used to control the Motor 4.
• depth: The depth value (in meters) measured by depth sensor.
• press: The pressure value (in Pa) measured by depth sensor.
• voltage: The voltage value (in V) of battery.
• roll: The roll angles (in degrees) measured by nine-axis IMU.
• pitch: The pitch angles (in degrees) measured by nine-axis IMU.
• yaw: The yaw angles (in degrees) measured by nine-axis IMU.
• a_x: The acceleration (in m/s
2
) along the x-axis in the body coordinate frame, measured by
nine-axis IMU.
• a_y: The acceleration (in m/s
2
) along the y-axis in the body coordinate frame, measured by
nine-axis IMU.
• a_z: The acceleration (in m/s
2
) along the z-axis in the body coordinate frame, measured by
nine-axis IMU.
• w_row: The angular velocity (in degrees/s) of rotation around the x-axis in the body coordi-
nate frame, measured by nine-axis IMU.
• w_pitch: The angular velocity (in degrees/s) of rotation around the y-axis in the body coor-
dinate frame, measured by nine-axis IMU.
• w_yaw: The angular velocity (in degrees/s) of rotation around the z-axis in the body coordi-
nate frame, measured by nine-axis IMU.
The models of the vehicle are developed in another work which may be published in the
journal of Ocean Engineering. The paper’s title is “Dynamic Modeling of Quadrotor AUV Using a
Novel CFD Simulation”.
Note : It is kindly remind that any paper that uses the Data should cite [2] and the above
paper if it is published.
Experimental Design, Materials and Methods
We used ‘Haizhe’ AUV as the experimental subject. Fig. 1 is the assembly diagram. Fig. 2 is
the actual prototype. ‘Haizhe’ is installed with 4 brushless motors (SUNNYSKY A2212 KV980 II),
4 propellers (Three-bladed Propeller Outer Diameter: 55 mm,Thread Pitch: 80mm), 4 electronic
speed control (HOBBYWING Skywalker 20A), 1 depth sensor (MS5803-01 BA), 1 nine-axis inertial
4 D. Ji, X. Yao and S. Li et al. / Data in Brief 39 (2021) 107477
Fig. 1. “Haizhe” assembly diagram.
Fig. 2. The prototype of ‘Haizhe’.
measurement unit (GY-MPU9250), and 1 microcontroller unit (STM32F407VET6). More detailed
information of main components is listed below:
Main Components Type
Microcontroller Un it STM32F407VET6
Brushless DC Motor SUNNYSKY A2212 KV980 II
Propeller (Counterclockwise) Three-blade paddle (Outer Diameter: 55 mm, Thread Pitch: 80 mm)
Propeller (Clockwise) Three-blade paddle (Outer Diameter: 55 mm, Thread Pitch: 80 mm)
Electronic Speed Control HOBBYWING SkyWalker 20A
Battery Sony Power Battery VTC5
Depth Sensor MS5803-01 BA
Nine Axis IMU GY-MPU9250
2.4G Wireless Module EBYTE E34 2G4D20D
SD Card Kingston 16 g SD Card
Figure 3 shows a complete data collection test of the ‘Haizhe’ AUV. Firstly, the fault type was
set and recorded for the ‘Haizhe’. Then, the initializer was executed to check whether each com-
ponent could work properly or not. After that, ‘Haizhe’ executed the main program and turned
on the function of data recording. And then ‘Haizhe’ began sailing underwater. It is worth not-
ing that different behavioral responses would be generated under the influence of different fault
types , but the main program did not change. After completing the main program, ‘Haizhe’ would
D. Ji, X. Yao and S. Li et al. / Data in Brief 39 (2021) 107477 5
Fig. 3. A complete data collection test of ‘Haizhe’.
stop data recording and automatically rose to the surface. Finally, the file system of ‘Haizhe’
would save the state data as a text file (also called a data sample).
We set five common fault types for ‘Haizhe’, including normal state, slight damage to the
propeller, severe damage to the propeller, failure of the depth sensor, and load increase. For
each fault type , ‘Haizhe’ was tested several times. For each time, ‘Haizhe’ ran the same program
and sailed underwater for 10–20 s to ensure that state data was long enough. The state data
recorded in each test were then used as a data sample, and the corresponding fault type was
the true label of the data sample.
In the experiment, ‘Haizhe’ had only one fault type at the same time, and there was no multi-
failure concurrency. The fault of depth sensor was not the hardware damage but artificially
added a bias item when reading the pressure value, which made the calculated depth value
deviate from the true value. For example, when the true depth was 0.5 m, the depth calculated
by the sensor was 0.6 m.
Ethics Statement
We declare that the manuscript adheres to Ethics in publishing standards and the submitted
dataset is the real data recorded in the experiment, and there is no act of stealing other people’s
data or modifying data.
Declaration of Competing Interest
We declare that we have no known competing financial interests or personal relationships
which have, or could be perceived to have, influenced the work reported in this article.
6 D. Ji, X. Yao and S. Li et al. / Data in Brief 39 (2021) 107477
CRediT Author Statement
Daxiong Ji: Conceptualization, Methodology; Xin Yao: Software, Data curation, Writing –
original draft; Shuo Li: Investigation; Yuangui Tang: Supervision; Yu Tian: Validation.
Acknowledgments
This study was supported by the Basic Public Welfare Research Plan of Zhejiang Province
(LGF 20E090 0 04 ).
References
[1] D. Ji, X. Yao , S. Li, Y. Tang, Y. Tian, Model-free fault diagnosis for autonomous underwater vehicles using sequence
convolutional neural network, Ocean Eng. 232 (2021) 108874, doi: 10.1016/j.oceaneng.2021.108874 .
[2] D. Ji, R. Wang, Y. Zhai, H. Gu, Dynamic modeling of quadrotor AUV using a novel CFD simulation, Ocean Eng. 237
(2021) 109651, doi: 10.1016/j.oceaneng.2021.109651 .
... To evaluate the effectiveness of the proposed HFAF network, experiments were conducted using the public dataset of the Haizhe AUV (Ji, Wang, et al. 2021). As can be seen in Figure 7 (Ji, Yao, et al. 2021), Haizhe AUV is a small quadrotor AUV developed in the laboratory, which is equipped with four brushless motors, four propellers, four electronic speed controllers, one depth sensor, one nineaxis inertial measurement unit (IMU) and one microcontroller unit. The specific equipment parameters were further elaborated in Ji, Yao, et al. (2021). ...
... As can be seen in Figure 7 (Ji, Yao, et al. 2021), Haizhe AUV is a small quadrotor AUV developed in the laboratory, which is equipped with four brushless motors, four propellers, four electronic speed controllers, one depth sensor, one nineaxis inertial measurement unit (IMU) and one microcontroller unit. The specific equipment parameters were further elaborated in Ji, Yao, et al. (2021). Moreover, regarding the dynamic model of the Haizhe AUV, it is nonlinear, strongly coupled, time-varying, multi-input and multi-output and includes unsteady payloads. ...
... We confirm that the data used in this study is publicly available and can be accessed through Ji, Yao, et al. (2021) [doi: 10.1016/j.dib.2021. Ethical and privacy concerns were not applicable to this study. ...
Hybrid feature adaptive fusion network for multivariate time series classification with application in AUV fault detection
Article
  • May 2023
Autonomous underwater vehicles (AUVs) acquire large-scale multivariate time series (MTS) data during navigation, which can be utilised to realise fault diagnosis, condition monitoring, and other functions by means of classifying the monitoring data. However, due to the complexity and time-variation of relationships between many variables of the MTS, we propose a MTS classification method, namely hybrid feature adaptive fusion network (HFAF). Specifically, a multi-scale method is first proposed to generate monitoring windows with different scales, and the spatiotemporal information is then fully obtained by dilated convolutional neural network (D-CNN) and dilated recurrent neural network (D-RNN). Subsequently, an adaptive feature fusion network based on an attention mechanism is introduced to address the redundancy and conflict between different scales. Finally, the hybrid feature network and adaptive fusion network are stacked up to form HFAF. The effectiveness and superiority of HFAF in AUV fault detection are demonstrated by the experiments conducted on Haizhe AUV, which yields more than 96% precision and more than 95% recall for various faults, outperforming other fault detection methods.
View
... The urgent needs in marine resource exploration in recent years have boosted the rapid development of underwater vehicles and relevant technologies, especially autonomous underwater vehicles (AUV). AUVs are free from entangled cables or support from the surface system and distinguished with wide range of activities, admirable diving depth, little requirement on deck area, low cost and the ability to access complex environment [1][2]. With the growing degree of autonomy, increasing numbers of control system as well as more eminent performance, AUVs have become capable with more and more complex tasks. ...
... With the models in Equations (1) and (2) At this point the AUV model is ( 1) ( , ( ), ( )) ...
STF-based diagnosis of AUV thruster faults
Article
Full-text available
  • Nov 2022
The diagnosis of thruster faults of autonomous underwater vehicles is studied in this paper. Based on the theory of strong tracking filter (STF), the AUV motion model and the thruster fault model are established. The STFs are designed for each thruster for the purpose of fault diagnosis. The AUV state and the fault deviation of the thruster are estimated online before the thruster faults are diagnosed based on residual analysis. The simulation experiments were conducted to verify the feasibility and effectiveness of the STF-based diagnosis of AUV thruster faults.
View
... The AUV dataset used in this paper, which includes possible fault types, was derived from the Zhejiang University laboratory and can be accessed at https://data.mendeley. com/datasets/7rp2pmr6mx (accessed on 28 June 2021) [30]. The AUV dataset comprises a total of 1225 samples, consisting of normal working conditions and four fault conditions. ...
Fault Diagnosis of Autonomous Underwater Vehicle with Missing Data Based on Multi-Channel Full Convolutional Neural Network
Article
Full-text available
  • Oct 2023
The fault feature extraction and diagnosis of autonomous underwater vehicles (AUVs) in complex environments pose significant challenges due to the intricate nature of the signals that reflect the AUVs’ states in the deep ocean. In this paper, an analytical model-free fault diagnosis algorithm based on a multi-channel full convolutional neural network (MC-FCNN) is introduced to establish patterns between AUV states and potential fault types using multi-sensor signals. Firstly, the AUV raw dataset undergoes random forest multiple imputation by chained equations (RF-MICE) to serve as the input of the convolution neural network. Next, signal features are extracted through the full convolution channel, which can be fused as multilayer perceptron (MLP) input and Softmax classifier for fault identification. Finally, to validate the effectiveness of the proposed MC-FCNN model, fault diagnosis experiments are conducted using the dataset sourced from the Zhejiang University Laboratory with missing data. The experimental results demonstrate that, even with 60% of the data missing, the proposed RF-MICE with MC-FCNN model can still achieve an ideal fault identification.
View
... To verify the validity of the proposed method MSF, we performed experiments using the public data set of Haizhe AUV 16 . The dataset contains 1225 data samples for 5 fault types and 17 monitoring variables. ...
A fault detection method for AUV based on multi-scale spatiotemporal feature fusion
Conference Paper
  • Mar 2023
Autonomous Underwater Vehicles (AUVs) are important equipment for ocean development and exploration. To ensure the task implementation of AUV, faults shall be detected in time. We propose a fault detection method based on Multiscale Spatiotemporal Feature fusion (MSF) for the time-varying characteristics and multiple correlation characteristics of AUV monitoring data. First, we apply a variety of sampling and data processing methods to generate monitoring windows with different scales along the time axis. Then, a composite feature extraction method is proposed to obtain temporal and spatial features simultaneously, and a feature pyramid of temporal and spatial information is formed. We use Bidirectional Long Short-Term Memory (BiLSTM) to obtain the time-series characteristics of a single monitoring variable, and Convolutional Neural Networks (CNN) to obtain the implicit spatial relationship characteristics among multiple monitoring variables. Next, we use an adaptive feature fusion method to solve the inconsistency in different feature scales, which can adaptively suppress the possible conflict information of different scale features. Finally, we use a fully connected network to detect the fault of the fused features. The fault detection experiment of Haizhe AUV shows the effectiveness and superiority of the proposed method.
View
View
View
Research on Multi-dimensional Time-series Data Fault Diagnosis Method for Autonomous Underwater Vehicles
Thesis
  • Nov 2023
With the continuous progress of marine science and engineering, Autonomous Underwater Vehicles (AUVs) have been playing an increasingly important role in underwater exploration and operations. They are widely used in various fields such as marine exploration, scientific research, environmental monitoring, resource development, as well as rescue and emergency response. AUVs can perform tasks in deep-sea and hazardous environments, thus reducing the safety risks faced by humans. To achieve autonomous navigation, environmental perception, and other functionalities, effective monitoring and analysis of the underwater environment are required. Multivariate time-series data, which describe the underwater environment and reflect the AUV's own motion and equipment conditions, are often obtained from multiple variables that change over time or captured by multiple sensors. To address the various risks of faults faced by AUVs during task execution, this study focuses on the diagnosis of AUV faults, specifically targeting multivariate time-series data. The research work of this paper focuses on fault detection, fault correlation analysis, the multi-source nature of monitoring data, and inter-domain differences. Specifically, this paper proposes the following research contents: (1) In the aspect of fault detection, a fault detection method based on an encoder-decoder architecture is proposed. Firstly, a multi-scale information enhancement method is introduced to generate monitoring windows of different scales, aiming to improve feature quality and model performance. Then, dilated convolutional neural networks and dilated recurrent neural networks are employed to learn the spatiotemporal information within each window. Subsequently, an attention-based adaptive fusion module is introduced to address redundancy and conflicts between different scales, enabling feature fusion across scales. Finally, the stacked hybrid feature module and adaptive fusion module form the encoder, which accomplishes the tasks of feature extraction and selection. To realize fault detection, a multilayer perceptron is utilized as the decoder to decode the fused features and obtain the operational status. Experimental results on a publicly available dataset for Haizhe AUV and a real-word dataset for Qianlong2 AUV fault detection demonstrate the effectiveness and superiority of the proposed approach. (2) In the aspect of fault correlation analysis, an attention-based fault correlation analysis method is proposed. This method first establishes a fault detection network to achieve basic detection performance. Then, time and spatial attention mechanisms are utilized to obtain the weight distribution of the model on monitoring variables, which quantitatively represents the correlation between faults and variables. To improve the quality of the weight distribution in the attention mechanisms, sparse constraints and feature orthogonalization methods are employed. The introduction of these methods effectively enhances the accuracy and robustness of the fault diagnosis model. Next, quantitative metrics for correlation analysis performance and a multi-task learning approach are introduced to enable the collaborative training of fault detection and correlation analysis tasks. By jointly learning the tasks of fault detection and correlation analysis, a more comprehensive understanding of fault characteristics and their correlation with monitoring variables is achieved, thereby improving the accuracy and interpretability of fault diagnosis. Finally, experiments conducted on the Qianlong-2 AUV validate the effectiveness of the proposed method. This approach provides new insights and methods for fault localization, traceability, and model interpretation in the fault diagnosis. (3) To address the hierarchical structure of AUV monitoring data, a hierarchical attention-based fault diagnosis method is proposed. Firstly, a comprehensive analysis of the hierarchical and multi-source characteristics of monitoring data is conducted, and actual faults are classified and defined. Next, a hierarchical attention-based fault diagnosis algorithm is proposed. This algorithm utilizes an encoder-decoder structure to perform multi-source fusion of monitoring data from the variable level to the device level and subsystem level, encoding features layer by layer. The fault correlation analysis method is applied to the attention mechanisms at different levels, enabling a hierarchical traceability approach based on attention weights. Furthermore, to address the issue of feature redundancy, a multi-head attention mechanism with random masking is introduced. The introduction of random masking improves the handling of redundant information in the multi-source fusion process, enhancing the accuracy and efficiency of fault diagnosis. Finally, experiments conducted on the Qianlong-2 AUV validate the powerful multi-source fusion capability of the proposed method, effectively achieving fault detection and fault traceability. The experimental results also demonstrate that the proposed method does not incur additional losses due to false positives during actual navigation, showcasing its potential and superiority in practical applications. (4) Based on the proposed fault detection and diagnosis methods, a generalized approach for model reuse in fault diagnosis models is proposed to address the issue of model domain differences in AUVs operating at different depths. This approach consists of two aspects: a generalized data generation method based on a multi-discriminator generative adversarial network and a multi-stage training strategy for generalized models. In the generalized data generation method, an adversarial generative network is employed to generate training data that covers different data distributions, and the introduction of multiple discriminators improves the generalization of the generated data. To ensure that the synthesized generalized data can capture the temporal dynamic characteristics of the original data, reconstruction loss and supervised loss for temporal dynamics are introduced to capture the temporal dynamic characteristics in time-series data. In the training strategy for generalized models, a multi-stage training approach is proposed. It utilizes generalized domain data, strong-labeled data, and weak-labeled data for modeling to reduce reliance on annotated data and construct a diagnostic model that generalizes to new operating depths. Multiple experiments conducted on the Haizhe AUV and Qianlong-2 AUV validate the feasibility and effectiveness of this method. As an approach to building a universal fault diagnosis model, this method holds practical value in the field of AUV.
View
View
View
Time-Efficient Neural Architecture Search for Autonomous Underwater Vehicle Fault Diagnosis
Article
  • Jan 2023
Autonomous underwater vehicle (AUV) can replace human to operate in complex underwater environment, so it must have the ability of self-fault diagnosis. Existing deep learning-based diagnostic methods have achieved excellent performance, but designing effective neural network structures is a time-consuming and difficult task. Although neural network architecture search (NAS) can automatically search effective neural network structures in a certain search space, NAS algorithms are usually slow and expensive. Therefore, this paper introduces a time-efficient NAS-based AUV fault diagnosis framework (TENAS-FD). TENAS-FD constructs a novel scoring algorithm that effectively gives a metric to characterize the performance of an untrained network. This metric is given based on the overlapping activation between data-points in the untrained network with different inputs. This allows TENAS-FD to search for superior network architectures in seconds on a single GPU. Experiments were conducted on a real AUV dataset and showed that TENAS-FD can quickly obtain excellent network architectures for AUV fault diagnosis and has better diagnostic performance compared to hand-designing models.
View
Dynamic modeling of quadrotor AUV using a novel CFD simulation
Article
  • Oct 2021
  • OCEAN ENG
Quadrotor AUV is a new and promising vehicle in applications due to its special advantages. However, the problem of dynamic modeling of quadrotor AUV has not been investigated yet. In this study, a new dynamic modeling method for a quadrotor AUV is proposed by using a novel CFD simulation to calculate the hydrodynamic coefficients. Star CCM + is used to simulate the uniform and variable motion of each degree of freedom of a quadrotor AUV to decouple original high coupling dynamic model. Through the simulation of the AUV and the propeller motion, the main hydrodynamic coefficients are calculated, while the mathematical relationship between the rotation velocity and the thrust/torque is concluded. Specifically, the static mesh and dynamic mesh models are used to simulate the uniform and variable motion of the quadrotor AUV respectively. The calculation of added mass and drag coefficient does not interfere with each other, leading to accurate results. A numerical method is also proposed to separate the linear and nonlinear parts of the total drag coefficients. Then a six DOF dynamic model of a quadrotor AUV is established. The proposed CFD simulation also provides a reliable, repeatable and low-cost method for dynamic modeling of underwater vehicles.
View
Model-free fault diagnosis for autonomous underwater vehicles using sequence Convolutional Neural Network
Article
  • Jul 2021
  • OCEAN ENG
The AUV must be capable of fault diagnosis if it is to perform tasks in complex environments without human assistance. However, the current fault diagnosis methods for AUV lack of manual experience and accuracy, leading to the lack of fault handling capacity. Different from the traditional model-based fault diagnosis, we propose a new model-free fault diagnosis method characterized by a deep learning-based algorithm, which is a new Sequence Convolutional Neural Network (SeqCNN) that learns the patterns between state data and fault type. More specifically, the proposed SeqCNN aims to extract global feature and local feature from state data and classify the extracted information into different fault types, and can convert two-stage diagnosis mode into a single-stage one. Compared to the traditional model-based diagnosis, it can significantly reduce the time-consuming burden, simplify the diagnosis procedure and improve the efficiency. The effectiveness of SeqCNN was validated by a practical experiment on a small quadrotor AUV ‘Haizhe’. The results indicate that the proposed SeqCNN can solve the problem of fault detection and fault isolation in single-stage diagnosis mode and that its accuracy is far superior to that of other deep learning diagnosis algorithms.
View