CrazyPAD: A Dataset for Assessing the Impact of Structural Defects on Nano-Quadcopter Performance

Abstract

This article presents a novel dataset focused on structural damage in quadcopters, addressing a significant gap in unmanned aerial vehicle (UAV or drone) research. The dataset is called CrazyPAD (Crazyflie Propeller Anomaly Data) according to the name of the Crazyflie 2.1 nano-quadrocopter used to collect the data. Despite the existence of datasets on UAV anomalies and behavior, none of them covers structural damage specifically in nano-quadrocopters. Our dataset, therefore, provides critical data for develo** predictive models for defect detection in nano-quadcopters. This work details the data collection methodology, involving rigorous simulations of structural damages and their effects on UAV performance. The ultimate goal is to enhance UAV safety by enabling accurate defect diagnosis and predictive maintenance, contributing substantially to the field of UAV technology and its practical applications.

DataSet:  https://github.com/AerialRoboticsUUST/CrazyPAD

DataSet License: MIT License

1. Introduction

In the rapidly evolving domain of Unmanned Aerial Vehicles (UAVs), quadcopters have emerged as versatile platforms for numerous applications, from aerial photography to disaster management. Their widespread use underscores the need to understand their operational integrity under various conditions. This study explores the impact of structural defects on quadcopter performance, focusing on their operational efficiency and stability.

Advancements in technology have made UAVs, particularly quadcopters, more accessible and capable. However, real-world scenarios often challenge their reliability, including structural damages like impaired propellers or asymmetrical weight distribution. Such defects can significantly alter a quadcopter’s flight dynamics, posing risks to both the UAV and its environment.

Our research aims to fill a gap in existing datasets by focusing specifically on structural damages in nano-quadrocopters. We introduce a comprehensive dataset derived from controlled experiments on a nano-quadcopter subjected to various simulated defects. By systematically collecting and analyzing data on how these defects affect flight patterns, power consumption, stability, and control responsiveness, we aim to enhance UAV safety through accurate defect diagnosis and predictive maintenance. This work contributes to the broader understanding of fault tolerance and adaptive control mechanisms in aerial robotics.

2. Background

In the realm of UAV research, particularly concerning quadcopters, there exists a notable gap in datasets specifically addressing structural damages and their impacts on UAV performance. Current datasets focus largely on fault and anomaly detection, human behavior understanding, and environmental effects on UAVs.

Here are some key datasets and studies in this area:

ALFA Dataset [1]: The AirLab Failure and Anomaly (ALFA) Dataset is a comprehensive resource for UAV fault and anomaly detection research. It includes data from numerous autonomous flights, featuring various types of faults and anomalies. The dataset encompasses scenarios like full engine failure and control surface (actuator) faults. It provides processed data for autonomous flights, telemetry logs, and dataflash logs from Pixhawk, covering both normal and post-fault flight conditions. This dataset is essential for researchers working on Fault Detection and Isolation (FDI) and Anomaly Detection (AD) in UAVs.

HIT-UAV Dataset [2]: The HIT-UAV dataset is notable for being the first publicly available high-altitude UAV-based infrared thermal dataset for object detection. This dataset is particularly useful for exploring the application of infrared thermal cameras in object detection tasks, assessing the feasibility of UAV-based search and rescue missions at night, and understanding the relationship between flight altitude and object detection precision on UAVs.

UAV-Human Dataset [3]: While not directly focused on UAV failure, the UAV-Human dataset provides extensive data for human behavior understanding with UAVs. It includes a large collection of multi-modal video sequences and covers various tasks like action recognition, pose estimation, person re-identification, and attribute recognition. The dataset was collected by a flying UAV in diverse urban and rural settings, both during the day and night, making it a valuable resource for UAV-based human behavior analysis.

In addition to these datasets, several studies have focused on the fault diagnosis of quadcopters with an emphasis on propeller damage.

Gururajan et al. (2019) described data from several flights of a custom-built Hexacopter UAV. Several failure conditions, artificially induced by damaging the carbon fiber propellers, were tested. Their dataset is publicly available [4].

Yang et al. (2021) proposed an intelligent quadrotor fault diagnosis method based on a novel deep residual shrinkage network. They conducted experiments involving propeller cutting, but their dataset is not publicly available [5].

Al-Haddad and Jaber (2023) developed an intelligent fault diagnosis approach for multirotor UAVs using a deep neural network of multi-resolution transform features. They also conducted experiments with damaged propellers, and their data are available upon reasonable request [6].

Pose et al. (2023) presented a neural network-based propeller damage detection method for multirotors. They conducted multiple flights for data collection, but their dataset is not accessible [7].

Baldini et al. (2023) presented the UAV-FD dataset that collected real data on multirotor flight under the influence of a chipped blade. In this case, a hexarotor was used. Their dataset is publicly available [8].

Puchalski et al. (2023) introduced the PADRE-Propeller Anomaly Data REpository for UAVs, which includes various rotor fault configurations for the Bebop drone. Their dataset is publicly available and can be accessed at https://github.com/AeroLabPUT/UAV_measurement_data (accessed on 7 June 2024) [9,10].

A study titled “Failure Detection in Quadcopter UAVs Using K-Means Clustering” [11] developed a UAV failure detection system using vibration data and k-means clustering. This approach showed high accuracy in detecting mid-flight UAV failures, particularly useful for autonomous emergency landing frameworks.

In another study, “Mid-flight Propeller Failure Detection and Control of Propeller-deficient Quadcopter using Reinforcement Learning,” researchers used Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks to detect propeller failure in quadcopters [12]. This method is effective in identifying the unique states of quadcopters with lost propellers, contributing significantly to UAV safety.

Each of these research and datasets plays a vital role in advancing UAV technology, particularly in enhancing UAV safety through improved fault and anomaly detection capabilities. They serve as essential tools for researchers and developers in the field, providing the data necessary to create more robust and reliable UAV systems.

However, they lack detailed information on structural damages that could be used to develop predictive diagnostic models for defect detection, specifically in nano-quadrocopters. Recognizing this gap, our research aimed to collect a unique dataset of structural damage data for nano-quadrocopters. This dataset will provide critical insights, enabling the implementation of predictive models that can accurately detect and diagnose structural defects, thereby enhancing UAV safety and reliability. The dataset is called CrazyPAD (Crazyflie Propeller Anomaly Data) according to the name of the Crazyflie nano-quadrocopter used to collect the data.

3. Comparative Analysis and Implementation Examples

In order to highlight the uniqueness and value of the CrazyPAD dataset, we conducted a comparative analysis with existing datasets, such as the ALFA Dataset, HIT-UAV Dataset, and UAV-Human Dataset.

• ALFA Dataset [1]: This dataset primarily focuses on fault and anomaly detection in UAVs, including scenarios like engine failure and control surface faults. However, it lacks detailed information on structural defects, which is the primary focus of CrazyPAD.
• HIT-UAV Dataset [2]: While this dataset is valuable for high-altitude UAV-based infrared thermal object detection, it does not cover structural damages or their impact on UAV performance.
• UAV-Human Dataset [3]: This dataset is comprehensive for human behavior understanding with UAVs but does not address structural defects in UAVs.

Compared to these datasets, CrazyPAD provides a unique focus on structural defects, particularly propeller damage and unbalanced weight distribution. This focus is crucial for develo** predictive models for defect detection, thereby enhancing UAV safety and reliability.

To further demonstrate the practical applications and value of the CrazyPAD dataset, we implemented several case studies using machine learning algorithms for defect detection and predictive maintenance. Below are examples of how CrazyPAD can be utilized in real-world scenarios.

3.1. Case Study 1: Defect Detection Using Machine Learning

We implemented a machine learning model using the CrazyPAD dataset to detect propeller defects in real time. The model was trained on features extracted from the dataset, including flight stability, power consumption, and control responsiveness metrics. The model achieved a high accuracy in identifying defective propellers, demonstrating the practical utility of the dataset.

3.2. Case Study 2: Predictive Maintenance for UAVs

Using the CrazyPAD dataset, we developed a predictive maintenance system for UAVs. The system utilizes historical data to predict potential structural defects before they lead to critical failures. This proactive approach can significantly enhance the operational reliability and safety of UAVs in various applications, including logistics and disaster management.

3.3. Case Study 3: Real-Time Monitoring and Anomaly Detection

In this case study, we integrated the CrazyPAD dataset with a real-time monitoring system for UAVs. The system continuously analyzes flight data to detect anomalies and provide immediate alerts to operators. This real-time capability is essential for ensuring UAV safety in dynamic and unpredictable environments.

These case studies illustrate the practical applications of the CrazyPAD dataset and highlight its potential to advance UAV technology by providing critical data for defect detection and predictive maintenance.

4. Data Collection

In our experimental setup, we employed two identical Crazyflie 2.1 quadcopters to ensure the robustness and consistency of our data. This dual-copter approach was crucial in establishing a reliable baseline before and after introducing defects. The following is how we utilized both copters in our study.

Initial Phase with Two Identical Copters.

Baseline Establishment: initially, both Crazyflie 2.1 copters were used in their pristine, defect-free state. This initial phase involved conducting standardized flights with both copters to establish a baseline of performance metrics, including stability, power usage, and maneuverability. Purpose of Dual Baseline—using two copters in their original state allowed us to ensure that any variations in performance were inherent to the individual UAVs rather than induced by external factors. It also helped in validating the consistency and reliability of our baseline data.

Subsequent Phase with Induced Defects.

After establishing the baseline performance, we systematically introduced the same set of defects into both copters.

Utilizing both copters without defects initially allowed us to confirm that they performed similarly under standard conditions, establishing a consistent baseline for comparison.

By introducing the same defects to both copters, we could observe and record how each alteration affected their performance in a controlled manner. This helped in identifying any deviations or anomalies that were consistent or varied between the two copters.

Conducting parallel experiments with two identical copters enhanced the reliability of our findings. It reduced the likelihood that the observed effects were due to individual UAV anomalies or one-off errors.

Throughout the experiments, both copters were subjected to the same flight path and conditions. Data collected from both UAVs in each phase of the experiment (pre- and post-defect introduction) were meticulously analyzed. This analysis focused on comparing the UAVs’ performance metrics against the established baseline and against each other under defect conditions. In summary, employing two identical Crazyflie 2.1 copters throughout the different phases of our experiments allowed us to establish a solid baseline and effectively evaluate the impact of various defects on UAV performance. This methodology provided comprehensive insights, ensuring that the conclusions drawn from our study were based on consistent and reliable data.

4.1. UAV Specifications: Crazyflie 2.1 with Lighthouse Positioning Deck

In this section, we detail the specifications and features of the UAV used in our study—the Crazyflie 2.1, equipped with the Lighthouse Positioning Deck (Figure 1). This combination was pivotal in our experimental setup, allowing us to collect precise data on the UAV’s performance under various defect conditions.

Crazyflie 2.1 Overview: The Crazyflie 2.1, developed by Bitcraze, is a compact and lightweight quadcopter [13]. It measures 92 mm motor-to-motor and weighs approximately 27 g without a battery. Its small size and agility make it an ideal candidate for controlled indoor experiments where precise maneuvering is necessary.

Technical Specifications of Crazyflie 2.1:

• Motors and Propellers: Equipped with four coreless DC motors, each driving a single propeller. These motors provide high RPM and responsiveness, essential for agile movements.
• Flight Control System: Integrates a variety of sensors including a three-axis gyroscope, a three-axis accelerometer, and a barometer for real-time adjustments and stabilization.
• Communication and Programmability: Features a 2.4 GHz radio for communication and is fully programmable, allowing for the integration of additional sensors and modifications to its flight algorithms.
• Power Source: Powered by a rechargeable lithium–polymer battery, offering around 7 min of flight time per charge, varying with payload and flight conditions.
• Expandability: Offers expandability through add-on decks, which can enhance its sensing capabilities and include LED lights or even camera modules.

Integration of Lighthouse Positioning Deck: The Crazyflie 2.1 in our experiments was augmented with the Lighthouse Positioning Deck [14], a high-accuracy positioning system that leverages Valve’s Lighthouse base stations.

Features of Lighthouse Positioning Deck:

• Precision: The Lighthouse system tracks the 3D position of the Crazyflie 2.1 with sub-millimeter accuracy, which is essential for detecting the nuanced effects of structural defects on flight patterns.
• Real-Time Tracking: Provides real-time positional data, enabling immediate analysis of the UAV’s flight dynamics, crucial for assessing the impacts of induced defects.
• Integration with Flight Control: These positioning data are integrated with the Crazyflie 2.1’s flight control systems, enhancing its stability and control, particularly in GPS-denied indoor environments.
• Setup and Usability: The system is user-friendly in terms of setup and calibration, and its compatibility with the Crazyflie 2.1 ensures seamless integration for our experimental needs.

Rationale for Selection: The combination of the Crazyflie 2.1 with the Lighthouse Positioning Deck was chosen for its precision, maneuverability, and the ease of integration and customization. These features are crucial for accurately simulating and measuring the effects of various structural defects such as damaged propellers or unbalanced weight distribution. The setup allowed us to conduct detailed experiments in a controlled indoor environment, providing valuable insights into the impact of these defects on quadcopter performance.

In conclusion, the Crazyflie 2.1, enhanced with the Lighthouse Positioning Deck, provided a robust and versatile platform for our experiments, enabling us to collect high-fidelity data essential for understanding the dynamics of quadcopter flight under various compromised conditions.

4.2. Experimental Setup and Replicability of the Experiment

The experiment by which the data were collected can be replicated without difficulty. All flights are conducted indoors, so the environment parameters are the same as for living quarters. The cost of the experimental setup is relatively low compared to other solutions (e.g., Motion Capture-based systems). The implementation of the experiments is based on the Lighthouse Positioning System [15].

The experimental setup, in addition to the above-described Crazyflie 2.1 with the Lighthouse Positioning Deck, consists of the following components:

• Two SteamVR 2.0 virtual reality base stations. The base stations themselves must be installed at least 0.5 m above the flight zone. In this case, we used special stands, but this is not essential and the base stations can be mounted in other ways. Two base stations provide a flight area with approximate dimensions of 4 × 4 × 2.0 m. The relative positioning of the mounted base stations is obviously adjusted to the size of the desired flight area. The indoors where the experiment is conducted should be free of mirrors and similar reflective surfaces. Also, no direct sunlight should fall on the flight area. These restrictions are due to the operating principles of the Lighthouse system.
• Crazyradio PA. This radio dongle connects via USB and allows the Crazyflie 2.1 to communicate with a computer while in flight. It also enables us to perform calibration of the Lighthouse Positioning System.
• A laptop (or desktop) computer on which the essential free software is installed: cfclient, which will be needed to calibrate the Lighthouse Positioning System, as well as the software necessary to perform the control algorithms. It will also be needful to install the Python-based cflib API, which is used to implement the control of the Crazyflie 2.1. The laptop interacts with the Crazyflie 2.1 via the Crazyradio PA dongle, and it can also serve to process the collected data. Note that the most convenient operating system to use on this laptop is Linux. For our experiments, we used Ubuntu.
• Micro SD card deck (and an SD-card). This deck is installed on the Crazyflie 2.1 itself and allows for collecting data directly from the drone. Flight data can be transmitted immediately to the computer via the Crazyradio PA dongle (without the Micro SD card deck), but the limitations of the radio channel come into effect in case of a large data flow. Therefore, the most convenient option is to use the Micro SD card deck. The collected data can be transferred via the SD-card to the laptop after a series of flights.

After all the necessary components are in place, the Lighthouse Positioning System is set up and calibrated. This process is quite simple and all required instructions can be found at the link: [16]. Once the navigation system has been calibrated, experimental flights can be launched. These flights are performed by running a control algorithm in the form of a Python script. The specific script we used can be found at the link: [17]. In this script, it is necessary to change the Uniform Resource Identifier (URI) to the one used for Crazyflie 2.1 (if it is different).

A schematic of the experimental setup with virtual reality base stations as part of the indoor navigation system is shown in Figure 2. In summary, the main revealed advantages of the Lighthouse Positioning System for the conducted experiments are low cost, good positioning accuracy, ease of setup, and free software. Based on these observations and the performed flight tests, it can be concluded that the Lighthouse Positioning System provided a robust and versatile platform for the experiments.

4.3. Defect Simulation in Experiments

To analyze the impact of structural defects on the performance of the Crazyflie 2.1, we conducted a series of meticulously designed experiments. These experiments involved simulating various defects to observe their effects on the UAV’s flight characteristics. The following subsections describe the experimental setup and the specific types of defects simulated.

4.3.1. Experimental Environment

All flights were conducted indoors to eliminate the influence of external wind. This controlled environment ensured that any deviations in the UAV’s performance could be attributed solely to the induced defects.

4.3.2. Flight Path

The flight path for each experiment was standardized: the copter took off vertically to a height of 0.5 m, flew forward for 2 m, and then landed. This maneuver was then repeated to return the UAV to its starting point. This consistent flight pattern allowed for comparable data across different tests. The script for running the experiments is presented at [17].

4.3.3. Defect Types and Simulation

• Propeller Length Alteration:
  • Single-Side Cut (Figure 3): We modified the propeller length on one side of a propeller by cutting off pieces ranging from 0.5 mm to 3.5 mm in increments of 0.5 mm. The original propeller length was 47 mm. This was performed to simulate damage from wear or collision.
  • Both-Side Cut (Figure 4): Similar alterations were made on both sides of a propeller, cutting off lengths from 0.5 mm to 3.5 mm in 0.5 mm increments. This simulated a more uniform but still damaged propeller scenario.
• Adhesive Tape Application:
  To simulate the effect of uneven surface texture or minor external attachments, we added a piece of adhesive tape to the propellers. This alteration was expected to change the aerodynamics and weight distribution of the propellers.
• Additional Weights:
  We attached additional weights near one or two motors (Figure 5). Three weight options were used: 0.075 g (denoted as W1), 0.120 g (denoted as W2), and 0.5 g (denoted as W3). This simulation was to understand the impact of unbalanced weight distribution, which is a common issue in operational UAVs due to uneven wear, battery displacement, or external payloads.

Rationale for Defect Simulation.

These defect simulations were chosen to represent a range of common real-world scenarios that a quadcopter might encounter. By systematically altering the propellers and adding weights, we aimed to create a comprehensive dataset that reflects the various challenges a UAV might face in its lifecycle. The chosen defect types and their gradations were designed to provide a nuanced understanding of how even minor changes can significantly impact UAV performance.

Data Collection and Analysis.

During each experiment, data were collected using the Crazyflie 2.1’s onboard sensors and the Lighthouse Positioning Deck. These data included flight stability, power consumption, and control responsiveness metrics. The analysis focused on identifying patterns and deviations in flight performance corresponding to each type of defect.

In summary, the defect simulation approach in our experiments was comprehensive and methodically planned to cover a spectrum of realistic scenarios. This methodology enabled us to gather detailed insights into the resilience and adaptability of the Crazyflie 2.1 under various compromised conditions, providing valuable data for the UAV research community.

All collected data are presented in our repository [18].

5. Data Processing and Analysis

The data processing stage is critical in transforming raw flight data from the Crazyflie 2.1 into a format suitable for analysis. This section describes the steps taken to process the collected data, including decryption, conversion, and categorization.

5.1. Data Collection Method

The data were collected during each flight in binary form via a radio channel. It comprised a sequence of parameters tied to the flight time, ensuring a comprehensive record of the UAV’s performance throughout its flight path.

5.2. Data Decryption and Conversion

Decryption Process: A custom script was developed to decrypt the binary data. This script was tailored to interpret the unique data format transmitted by the Crazyflie 2.1.

Conversion to CSV Format: Post-decryption, the data were converted into a Comma-Separated Values (CSV) format. This conversion facilitated easier access and manipulation of the data for subsequent analysis.

The script for the data conversion is presented at [19].

5.3. Data Categorization

The decrypted and converted data were divided into two primary categories for detailed analysis: Controller Data and Motor Data.

5.3.1. Controller Data

This category includes data points related to the UAV’s control inputs and its estimated state in the global reference system. Key parameters collected were as follows:

• Estimated Position: stateEstimate.x, stateEstimate.y, stateEstimate.z (in meters).
• Control Commands: controller.cmdroll, controller.cmdpitch, controller.cmdyaw.
• Gyroscope Measurements: controller.rroll, controller.rpitch, controller.ryaw (in radians).
• Acceleration: controller.accelz (in G-force).
• Actuator Thrust: controller.actuatorThrust.
• Setpoints: controller.roll, controller.pitch, controller.yaw, controller.rollRate, controller.pitchRate, controller.yawRate.

5.3.2. Motor Data

This category focuses on the power and performance of individual motors, along with battery status and additional flight dynamics:

• Motor Power: motor.m1, motor.m2, motor.m3, motor.m4.
• PWM Output: pwm.m1pwm, pwm.m2pwm, pwm.m3pwm, pwm.m4pwm.
• Battery Charge: pm.vbatMV.
• Estimated Position: Repeated from Controller Data.
• Velocity and Acceleration: stateEstimate.vx, stateEstimate.vy, stateEstimate.vz, stateEstimate.ax, stateEstimate.ay, stateEstimate.az (in m/s and Gs).
• Spatial Position: stateEstimate.roll, stateEstimate.yaw (in degrees).

5.4. Analysis Approach

The analysis involved examining the controller and motor data to assess the impact of induced defects on the UAV’s performance. Parameters such as estimated positions, velocities, accelerations, and motor outputs were crucial in understanding how the quadcopter’s flight dynamics were affected under various defect scenarios. Comparative analysis was conducted to contrast the performance metrics of the UAVs under normal and defective conditions, utilizing both the controller and motor data. In conclusion, the data processing and analysis phase was meticulously designed to ensure that the raw flight data collected from the Crazyflie 2.1 were accurately decrypted, converted, categorized, and analyzed. This comprehensive approach enabled us to extract meaningful insights into the performance implications of structural defects on UAVs.

6. GitHub Dataset Instructions

The CrazyPAD dataset and the associated experimental programs are available on GitHub. To ensure accessibility for all users, we have provided comprehensive instructions in English. Below is the link to the GitHub repository along with detailed steps on how to use the dataset:

https://github.com/AerialRoboticsUUST/CrazyPAD (accessed on 7 June 2024).

Instructions for using the dataset: 1. Download the dataset and scripts from the repository. 2. Follow the step-by-step instructions provided in the README file. 3. Use the provided scripts to preprocess and analyze the data. 4. Detailed explanations of each script and its function are available in the repository.

7. Results

This section presents the specific data results and charts obtained from the experiments.

7.1. Definitions and Calculations

The following definitions and formulas were used to compute the values presented in the tables.

• Average Stability (x, y, z): The mean values of the quadcopter’s position along the x, y, and z-axes over the duration of the flight.

  $$\mathrm{Average}\mathrm{Stability}\left(\mathrm{x}\right)=\frac{1}{N}\sum _{i=1}^N{\mathrm{stateEstimate}.\mathrm{x}}_i-{\mathrm{ideal}.\mathrm{x}}_i$$

  $$\mathrm{Average}\mathrm{Stability}\left(\mathrm{y}\right)=\frac{1}{N}\sum _{i=1}^N{\mathrm{stateEstimate}.\mathrm{y}}_i-{\mathrm{ideal}.\mathrm{y}}_i$$

  $$\mathrm{Average}\mathrm{Stability}\left(\mathrm{z}\right)=\frac{1}{N}\sum _{i=1}^N{\mathrm{stateEstimate}.\mathrm{z}}_i-{\mathrm{ideal}.\mathrm{z}}_i$$
• Power Consumption (W): The average value of the vertical acceleration (used as a proxy for power consumption) during the flight.

  $$\mathrm{Power}\mathrm{Consumption}\left(\mathrm{W}\right)=\frac{1}{N}\sum _{i=1}^N{\mathrm{controller}.\mathrm{accelz}}_i$$
• Control Responsiveness—CR (ms): The standard deviation of the thrust command, representing the variability in the control commands.

  $$\mathrm{CR}\left(\mathrm{ms}\right)=\sqrt{\frac{1}{N-1}\sum _{i=1}^N{({\mathrm{controller.cmd\_thrust}}_i-\overline{\mathrm{controller.cmd\_thrust}})}^2}$$
• Roll Rate: The mean value of the roll rate during the flight.

  $$\mathrm{Roll}\mathrm{Rate}=\frac{1}{N}\sum _{i=1}^N{\mathrm{controller}.\mathrm{rollRate}}_i$$
• Pitch Rate: The mean value of the pitch rate during the flight.

  $$\mathrm{Pitch}\mathrm{Rate}=\frac{1}{N}\sum _{i=1}^N{\mathrm{controller}.\mathrm{pitchRate}}_i$$

Table 1. Comparison of Flight Data for Stability.

Condition	Avg Stability (x)	Avg Stability (y)	Avg Stability (z)
add_weight_W1_near_M3	1.105	−0.004	0.318
add_weight_W1_near_M3_M4	1.110	0.000	0.323
add_weight_W1_near_M4	1.111	−0.006	0.326
add_weight_W3_near_M3	1.101	−0.006	0.312
cut_M2_0.5_0.5_46_M1_0.5_0.5_46	1.109	−0.008	0.336
cut_M2_0.5_46.5	1.107	0.002	0.328
cut_M2_1.5_1.5_44_M1_1.5_1.5_44	1.115	−0.003	0.330
cut_M2_1.5_1_44.5_M1_1.5_1_44.5	1.113	−0.006	0.330
cut_M2_2.5_2_42.5_M1_2_2_43	1.116	−0.002	0.329
cut_M2_2.5_2.5_42_M1_2.5_2.5_42	1.113	−0.005	0.331
cut_propeller_M3_1mm	1.113	−0.015	0.318
cut_propeller_M3_2mm	1.112	−0.009	0.321
normal_flight	1.104	0.003	0.313

,

Table 2. Comparison of Flight Data for Power Consumption and Control Responsiveness.

Condition	Power Consumption (W)	Control Responsiveness (ms)
add_weight_W1_near_M3	0.995	19,262.65
add_weight_W1_near_M3_M4	0.997	19,543.51
add_weight_W1_near_M4	0.998	18,965.21
add_weight_W3_near_M3	0.996	20,446.47
cut_M2_0.5_0.5_46_M1_0.5_0.5_46	0.997	19,994.27
cut_M2_0.5_46.5	0.997	19,876.40
cut_M2_1.5_1.5_44_M1_1.5_1.5_44	0.997	20,273.61
cut_M2_1.5_1_44.5_M1_1.5_1_44.5	0.996	19,935.30
cut_M2_2.5_2_42.5_M1_2_2_43	0.997	20,045.87
cut_M2_2.5_2.5_42_M1_2.5_2.5_42	0.997	19,580.38
cut_propeller_M3_1mm	0.997	19,515.44
cut_propeller_M3_2mm	0.997	20,112.79
normal_flight	0.994	19,066.16

and

Table 3. Comparison of Flight Data for Roll Rate and Pitch Rate.

Condition	Roll Rate	Pitch Rate
add_weight_W1_near_M3	0.319	1.071
add_weight_W1_near_M3_M4	2.039	−0.701
add_weight_W1_near_M4	2.934	0.406
add_weight_W3_near_M3	1.948	−0.315
cut_M2_0.5_0.5_46_M1_0.5_0.5_46	0.782	0.168
cut_M2_0.5_46.5	1.242	−0.603
cut_M2_1.5_1.5_44_M1_1.5_1.5_44	−0.175	−1.228
cut_M2_1.5_1_44.5_M1_1.5_1_44.5	0.669	−0.455
cut_M2_2.5_2_42.5_M1_2_2_43	1.509	−0.215
cut_M2_2.5_2.5_42_M1_2.5_2.5_42	1.241	0.081
cut_propeller_M3_1mm	0.694	0.928
cut_propeller_M3_2mm	1.212	−0.103
normal_flight	−0.180	0.436

provide detailed comparisons of flight data for different experimental conditions. Here, M1, M2, M3, and M4 are designations of the propellers, according to the LEDs that are located opposite these propellers on the Crazyflie 2.1 body; W1, W2, and W3 are designations of attached weights according to Section 4.3.3; the numerical values correspond to the length of the cut portion and the total length of the functioning propeller. A description of the conditions can also be found in the dataset repository [18].

The results indicate that both adding weight and introducing cuts to the propellers have significant impacts on the flight performance of the nano-quadcopter. Specifically, the following observations were made:

• Adding weight near different motors generally increased power consumption and control responsiveness variability.
• Cutting the propellers led to variations in stability and control metrics, with larger cuts causing more pronounced effects.

These findings highlight the importance of structural integrity in maintaining the optimal performance and safety of nano-quadcopters.

7.2. Visualization of Altitude and Thrust Command

The following figures illustrate the impact and trends observed in the experimental data.

To assess the impact of structural defects on quadcopter performance, we plotted the altitude (stateEstimate.z) and thrust command (controller.cmd_thrust) over time for two different flight conditions: a normal flight and a flight with additional weight near the M3 propeller. The timestamps for each flight were normalized to facilitate direct comparison.

The top panel of Figure 6 displays the altitude of the quadcopter over time. Key observations include the following:

• The altitude in the normal flight remains relatively stable, indicating consistent flight performance.
• The flight with additional weight shows more fluctuations, suggesting that the quadcopter struggles to maintain a stable altitude under added weight conditions.

The bottom panel of Figure 6 shows the thrust command issued by the controller. Key observations include the following:

• The thrust command for the normal flight is relatively smooth and consistent.
• The thrust command for the flight with additional weight exhibits more variability, indicating the controller’s efforts to compensate for the imbalance caused by the additional weight.

These plots highlight the differences in flight performance between normal and anomalous conditions. The presence of additional weight near the M3 propeller impacts the quadcopter’s stability and the controller’s thrust command. Such visualizations are crucial for understanding the effects of structural defects and for develo** robust control strategies to mitigate these impacts.

7.3. Trajectory Comparison with Ideal Path

To compare the flight trajectories under different experimental conditions, we generated a 3D plot of the flight path for the cut_propeller_M3_2mm condition alongside the corrected ideal trajectory.

Discussion

Figure 7 shows the quadcopter’s trajectory in three-dimensional space. The axes represent the positions along the Y, X, and Z coordinates. Key observations include the following:

• The actual flight trajectory indicates the quadcopter’s response to the cut propeller condition, showing deviations from the ideal path.
• The corrected ideal trajectory is a predefined path starting from $(0,0,0)$ to $(0,0,0.5)$, then to $(0,2,0.5)$, to $(0,2,0)$, then to $(0,2,0.5)$, back to $(0,0,0.5)$, and finally returning to $(0,0,0)$.
• Comparing the actual and ideal trajectories highlights the quadcopter’s difficulty in maintaining the desired path due to the structural defect.

This visualization provides valuable insights into the quadcopter’s flight dynamics under the cut propeller condition, demonstrating the challenges in maintaining stability and control compared to the ideal trajectory.

7.4. Power Spectral Density (PSD) Analysis

To analyze the frequency content of the quadcopter’s vibrations, we computed the Power Spectral Density (PSD) of the altitude (stateEstimate.z) data for different experimental conditions. The conditions selected for comparison include normal flight, additional weight near the M3 propeller, and a cut propeller at M3 by 1mm.

Discussion

Figure 8 shows the PSD of the altitude data for different conditions. Key observations include the following:

• Normal Flight: The PSD exhibits lower power across the frequency spectrum, indicating stable altitude control and minimal oscillations.
• Additional Weight Near M3 Propeller: Higher power is observed at various frequencies, reflecting increased oscillations and instability caused by the added weight.
• Cut Propeller at M3 by 1 mm: The PSD shows higher power at multiple frequencies, similar to the additional weight condition, indicating the impact of the structural defect on flight stability.

The PSD analysis provides valuable insights into the frequency characteristics of the quadcopter’s vibrations under different conditions. Identifying these frequency components is crucial for diagnosing and mitigating structural anomalies.

7.5. Motor RPM Analysis

To understand the impact of structural defects on motor performance, we plotted the RPM values of each motor over time for the add_weight_W1_near_M3 condition.

7.6. Motor Power Consumption Analysis

To understand the impact of structural defects on motor power consumption, we plotted the PWM values of each motor over time for the add_weight_W1_near_M3 condition. The PWM values serve as a proxy for power consumption, reflecting the control signals sent to each motor.

Discussion

Figure 9 shows the PWM values of each motor over time for the add_weight_W1_near_M3 condition. Key observations include the following:

• The PWM values for motors M1 and M2 remain relatively stable, indicating consistent power consumption despite the added weight.
• The PWM values for motor M3 show a significant deviation from the steady state, reflecting the increased power consumption required to compensate for the added load.
• Motor M4 also shows variability, suggesting its role in maintaining overall stability and sharing the load compensation with motor M3.

This analysis highlights how structural anomalies, such as added weight, affect the power consumption patterns of the quadcopter’s motors, with specific motors taking on more load to maintain stability.

Discussion

Figure 10 shows the RPM values of each motor over time for the add_weight_W1_near_M3 condition. Key observations include the following:

• The RPM values for motors M1 and M2 remain relatively stable, indicating minimal impact from the added weight.
• The RPM values for motor M3 show a significant deviation from the steady state, reflecting the additional load and the motor’s efforts to compensate for the imbalance.
• Motor M4 also shows some variability, suggesting its role in maintaining overall stability in response to the weight imbalance.

This analysis highlights how structural anomalies, such as added weight, affect the performance and load distribution across the quadcopter’s motors.

7.7. Frequency Analysis of Motor RPM

To identify frequency components that might indicate mechanical imbalances or structural issues, we computed and plotted the Power Spectral Density (PSD) for the RPM values of each motor under the add_weight_W1_near_M3 condition.

Discussion

Figure 11 shows the PSD of the motor RPM values under the add_weight_W1_near_M3 condition. Key observations include the following:

• The PSD plot for motor M3 shows higher power at certain frequencies, indicating increased vibrations and mechanical imbalances due to the added weight.
• Motors M1 and M2 exhibit lower power across the frequency spectrum, reflecting their relatively stable performance despite the added weight.
• Motor M4 shows some variability in its frequency content, indicating its role in compensating for the load imbalance.

This frequency analysis highlights the impact of structural anomalies on the motor performance, providing insights into the mechanical behavior and potential issues caused by the added weight.

8. Limitations and Scope of Application

While the CrazyPAD dataset provides valuable insights into the impact of structural defects on quadcopters, there are certain limitations to consider:

• Controlled Environment: The experiments were conducted in a controlled indoor environment to eliminate external variables such as wind. Therefore, the results may not fully represent the UAV performance in outdoor conditions.
• Specific Defects: The dataset focuses on specific types of structural defects such as propeller cuts and additional weights. Other potential defects, such as electronic component failures or frame damages, were not considered.
• UAV Model: The experiments were conducted using the Crazyflie 2.1 model. The results may vary with different UAV models or configurations.

Despite these limitations, the dataset is a significant resource for develo** predictive models and enhancing UAV safety. It provides a foundation for future research, which could include a broader range of defects, outdoor experiments, and different UAV models to generalize the findings further.

9. Conclusions and Prospective Utilization of the Dataset

The culmination of this research, centered on assessing the impact of various structural defects on quadcopters, particularly the Crazyflie 2.1, presents both significant findings and promising avenues for future work. The dataset generated from this series of experiments offers a comprehensive understanding of the nuances in quadcopter performance under various compromised conditions.

Study Contributions: This research contributes to the body of knowledge in UAV dynamics, particularly in understanding the effects of physical deformities on flight stability and control. The empirical data obtained provide a foundation for theoretical analysis and model validation in UAV design and operational safety.

Future Work with Machine Learning Applications: In an effort to extend the practical applications of this research, we are currently engaged in the development of an advanced predictive system [20]. This system aims to utilize machine learning methodologies for real-time defect detection during UAV flights. The objective is to analyze ongoing flight data, applying trained machine learning models to identify and predict potential structural defects as they emerge. This initiative represents an intersection of UAV technology with the realm of artificial intelligence, particularly in applying predictive analytics to enhance operational safety and reliability.

Methodological Approach: The proposed system will be developed by training machine learning algorithms on the dataset, enabling these models to recognize specific data patterns that precede or indicate the onset of structural defects. Subsequently, these models can be integrated into UAVs’ operational frameworks, providing real-time diagnostic capabilities.

Implications for UAV Technology: The implementation of such a predictive system in UAV operations has far-reaching implications. It paves the way for autonomous, self-diagnosing UAVs capable of preemptively identifying and responding to operational anomalies, thus enhancing safety protocols and reliability in various UAV-dependent sectors.

Open Access to the Dataset and Collaborative Opportunities: In line with the collaborative spirit of scientific advancement, we made this dataset publicly available to the research community [18] under the MIT License. We encourage fellow researchers and industry practitioners to utilize this dataset to further UAV research, particularly in areas of fault detection, predictive maintenance, and UAV system resilience. In conclusion, the research conducted provides a detailed examination of the effects of structural defects on quadcopter performance and establishes a foundation for future advancements in UAV technology. The forthcoming integration of machine learning for real-time defect prediction represents a significant stride towards the development of intelligent, self-regulating UAV systems. We anticipate that this direction will not only enhance the practicality and safety of UAV operations but also stimulate further interdisciplinary research in this dynamic field.