Energy harvesting and ANFIS modeling of a PVDF/ GO-ZNO piezoelectric nanogenerator on a UAV

: The aim of this study is to energy harvesting and modeling from a soft and ﬂ exible nanogenerator based on nano ﬁ bers (PVDF/ZnO-RGO) on the arm of a multi-rotor UAV. The nanogenerator modeling was done to improve the usability of piezoelectric nanogenerators to provide part of the electric energy required by drones. Tests were conducted on the experimental setup and practical ﬂ ight conditions. In each test and a period of 30 s, the vibration signals of the UAV arm were stored, and the vibration data extracted in the time domain and the electrical voltage of the nanogenerator were obtained through the vibrations of the drone, and its values were recorded in each experiment. After signal processing, the best vibration features were used for modeling. In the modeling process, some data were considered for training and testing in ANFIS network. The ANFIS inputs included the selected vibration features and the outputs included the stored voltage. The modeling results were analyzed based on increasing the amplitude of vibrations and the duration of exposure of the nanogenerator to vibration led to higher voltage generation. The maximum voltage recorded during the 30-s ﬂ ight time of the tested UAV was 64 mV at 3.54, 49.8, and 0.48 Hz along the x , y, and z axes, respectively. The results showed that the application of energy harvesting on di ﬀ erent vibration systems could be improved through modeling. It can be used to predict the amount of electric energy produced according to the vibrations of the drone in the future.


Introduction
Today, researchers are interested in renewable energy harvesting technologies for compact and portable electronic devices.The reduction of fossil fuels, on the one hand, a short life span that requires frequent charging and replacing batteries, creates limitations for electronic devices [1].Energy harvesting through nanomaterials and using sustainable energy sources and optimization of piezoelectric can overcome that limitation [2].Recently, the use of renewable energy has increased through the surrounding stimuli and alternative sources [3][4][5].Piezoelectric materials are lightweight, sensitive, stable, and with wide bandwidth and fast response [6].Piezoelectric transducers are good candidates for generating electrical energy by vibrations [7].Piezoelectric nanogenerators are significantly attractive [8].The piezoelectric material polyvinylidene fluoride material (PVDF) is one of the most wide polymers [9][10][11].Due to their high surface-to-volume ratio, nanomaterials are suitable for piezoelectric harvesters if subjected to deformation and vibration.The operating system of drones (Unmanned Aerial Vehicles) has a limited amount of electricity to provide electricity in order to fly for a fixed time period [12,13].Increasing the number of UAV batteries increases the total mass and generates more energy.The use of light and cheap piezoelectric nanogenerators to provide a small part of the electrical energy required by the drone can be a suitable option and piezoelectric application is helpful in low-power wireless electronic sensors in UAVs [14].Adjusting and optimizing energy harvesting devices based on piezoelectric vibration is essential, and various methods have been developed and check piezoelectric voltage, including experimental methods, mathematical modeling, and simulation [15][16][17].So, considering the importance of using renewable and clean energies, the limitation of batteries, as well as the limitation of excess weight on the UAVs, and on the other hand, the complexity of determining the exact amount of piezoelectric production voltage on the UAVs, the use of light piezoelectric nanogenerator and its modeling to predict and optimize energy harvesting on a UAVs is essential.The application of methods based on soft calculations and fuzzy has been presented [18][19][20].In a study after the software modeling of a quad-copter drone, four pieces of piezoelectric accelerometer were installed on the arm of the drone, and modal analysis was performed in laboratory conditions.The drone was stimulated by a shaker and a complete dynamic analysis of the UAV, and the electrical power levels generated by the UAV vibrations during the laboratory hover flight was performed [21].
Increasing the effective or operating frequency bandwidth based on piezoelectric vibration of energy harvesters is important to provide more resonances and generate more electric power.Various approaches have been proposed to further maximize the effective frequency bandwidth.In one paper, a mass was elastically connected to the end of the cantilever tip with the help of a spring.In order to optimize the energy harvester of the cantileverspring-oscillator system using soft calculations (genetic algorithm and Adaptive Neural Fuzzy Inference System), stiffness ratios were considered as input and resonance frequencies as output of the model.A total of 64 rules were created from the 8 × 8 fuzzy structure.The effect of restoring force and inertia on the dynamic and electrical behavior of the console was investigated which increased the voltage based on the results of the blows, and then, it was optimized for the highest efficiency by the soft calculation method [18].Fuzzy models include fuzzy rules formed by a number of fuzzy or linguistic values, allowing the researcher to understand the role of each input in the model compared to other inputs and outputs.Therefore, a piezoelectric actuator can be modeled using fuzzy subtractive clustering and fuzzy neural networks [22].In order to develop an Internet of Things (IoT)-based system to generate electricity for total power generation, some renewable energy sources were used, such as multiple sensors and two different Artificial Intelligence models such as Artificial Neural Network (ANN), and ANFIS [23].Today, precision agriculture using sensors to monitoring and agricultural applications [24] and as well as digital twin technology can be used to duplicate processes in other to predict their performance [25].In research aimed at improving the electromechanical properties of the nanogenerator, numerical modeling and optimization of the oxide-based piezoelectric nanogenerator were performed [2].In a comparative study of maximum power point tracking techniques from hybrid renewable energy systems using Fuzzy neural systems, ANFIS was mentioned among the best methods.An adaptive neural fuzzy inference system to predict the effect of four independent variables, including initial pollutant concentration, PH, light irradiation intensity, and time reaction, on the photocatalytic performance of Ag-TiO2/PVDF-HFP composite membrane on MNZ degradation in water treatment and sewage was used.The high R 2 value of 0.98 showed that the model was developed appropriately [26].Regarding the modeling and simulation of an open channel PEHF system for effective energy harvesting of PVDF, the output obtained using experiments was in good agreement with the results retrieved through simulation and mathematical results and ANFIS modeling [27].The fuzzy inference system based on the adaptive neural system successfully controlled and predicted the state of the artificial human finger using a wearable device and a pressure sensor [9].Active multi-mode vibration control of a smart structure using fiber composite macro actuators with feedback from PVDF membrane sensors used the ANFIS.The results showed that an ANFIS controller is the best choice in the conditions of the experiment [6].The use of flexible and light nanoparticles PVDF/ZnO-RGO to provide a small part of the electrical energy required by the UAVs is one of the innovative aspects of this study.
In this study, an attempt has been made to harvest energy from the PVDF used in the nanogenerator using the vibrations of a multi-rotor with agricultural applications.ANFIS model was used to analyze the effect of vibration intensity on extracted electrical energy.The stages of this study included the construction of the nanogenerator, vibration data collection, signal processing, electrical energy harvesting, and ANFIS modeling.

Materials and methods
This study with the aim of energy harvesting and modeling from a soft and flexible nanogenerator based on nanofibers (PVDF/ZnO-RGO) was conducted on an agricultural spraying drone in 2023 in Kermanshah, Iran.The stages of this study included the construction of the nanogenerator, vibration data collection, signal processing, electrical energy harvesting, and ANFIS modeling as shown in Figure 1.

ANFIS modeling
In this study, according to Figure 2, flexible and light nanogenerators were made based on PVDF/Zno-RGO polymer nanofibers through the electrospinning process [28].Then, by checking the amount of output voltage from the nanogenerator, the analysis of the amount of energy harvesting and the feasibility of its use in multi-rotor UAVs were discussed.Then, the ANFIS, which is the combination of fuzzy systems based on logical rules and the method of artificial neural networks, was used for modeling.The nanoparticles from US Research and hydrochloric acid (HCl) manufactured by Merck.

Nanogenerator construction
Since it was intended to prepare a thin and lightweight nanogenerator for use in drones, an electrospinning polymer solution, a cheap and straightforward method, was used to produce fragile fibers.The materials used to make the nanogenerator are given in the chart of Figure 2. The polyvinylidene fluoride was from Kynar, zinc oxide (ZnO) nanoparticles from US Research, and hydrochloric acid manufactured by Merck.

PVDF/Zno-RGO nanofiber structure
The first to make graphene oxide (GO), reduced graphene oxide (RGO), and nanocomposite (ZnO-RGO), RGO, and ZnO nanoparticles were used.Then, nanocomposite (ZnO-RGO) with PVDF polymer was used to prepare the polymer solution.The manufacturing of PVDF/ZnO-RGO nanopolymer is shown in Figure 3.
The experimental setup of nanogenerator-making tools used is shown in Figure 4.In this study, GO was made by the modified Hamers method [15].GO was reduced to eliminate functional groups and reduce structural defects.Ascorbic acid was a reducing agent.Then RGO was constructed.The mass measurements of the materials were done with a digital scale with an accuracy of 0.0001 mg.The centrifugal device used was the Universal model.A German model IKA RH basic magnetic stirrer was used to prepare nanocomposite (ZnO-RGO).In the preparation of the nanocomposite, the drying temperature of 100°C was selected through the Memmert model oven.Ultrasonic waves were used to homogenize the solution.
The amount of nanofiber feeding by the injection syringe and the distance between the sprayer and the collector affect the quality of the produced nanofibers.The feeding rate of the polymer solution was 0.4 ml per hour, and the distance between the sprayer and the collector was fixed at 20 cm.The optimal condition of the initial parameters to reach the maximum voltage of the nanogenerator was done through trial and error.If the viscosity is favorable, increasing the concentration of the polymer solution will increase the output voltage of the nanogenerator to some extent.RGO was combined with ZnO to produce light, flexible, and high piezoelectric nanofibers.ZnO has both semiconductor and piezoelectric properties.Using it to coat RGO oxide will significantly improve the piezoelectric property of PVDF polymer.Increasing the concentration of nanoparticles also increases the output Energy harvesting and ANFIS modeling  3 voltage, improves the piezoelectric properties, and positively increases the β phase.The ratio of nanoparticles should be such to improve the piezoelectric properties of PVDF fibers while preventing the increase in the thickness of nanofibers.The polymer solution concentration and nanoparticle concentration were 12.50 and 5.5%, respectively.The ratio of nanoparticles (ZnO-RGO) is 1.75 in the electrospinning process, and increasing the applied voltage increases the electrostatic forces in the Taylorized cone of more stable nanofibers.The electrospinning voltage was 17 kV and at room temperature and 23% humidity.

Construction of nanogenerator
The active material of nanogenerators consisted of two thin films of copper electrodes with size 1 × 4 × 0.5 with a piezoelectric layer in between and in the size of nanofibers.Electrodes on both sides of the nanogenerator transferred electricity to the circuit.In order to prevent the separation of nanofibers on the electrodes, paper glue was used.The glue prevented short-circuiting of the two electrode ends.Considering the drone's aerodynamics, the piezoelectric can be installed inside, on, or under the arms of the drones.As shown in Figure 5, the piezoelectric nanogenerator was installed on the arm of the UAV in the middle part where there was the most strain, integrally with the arm.The nanogenerator was closed by a tight glue and integrated into the drone arm to receive the arm's vibrations.

Vibration data collection
The prepared test set is shown in Figure 6.Experiments under vibration and impact on the nanogenerator were carried out under different conditions, and the results are given by the authors in previous research [28].These study experiments were carried out in real flight conditions on an agricultural sprayer hexacopter drone.The tested drone was made in China, Joyance Company, HT10L-606 model.This UAV was with 10.5 kg self-mass, two pcs batteries 17,000 mAh, 10-15 min fly time, and folded size of 0.64 M, 1.14 M width, and 0.58 M height.Vibration energy harvesters should be able to respond to low frequency and low acceleration vibrations that are usually present in the environment.On the other hand, due to the structure of the chassis, which is usually made of hard fiber carbon, UAVs need equipment with low vibration, so they have vibrations with low amplitude and frequency.In order to detect the vibration levels, the capabilities of smartphones and programs suitable for vibration measurement were used.
In this study, the Poco m3 smartphone was used with the vibration meter toolbox under the Android I Dynamic application [29].Sensor specifications were version 15,933, power 0.15 mA, resolution 0023942017 M/s 2 , and maximum range 78.4532 M/s 2 .The corresponding application saved the vibration data in CSV format, and after the end of the tests, it was transferred to the personal computer.The distribution of weight on sprayer drones varies with the consumption of spraying solution; therefore, despite the turbulence of the solution inside the tank and the decrease or increase of weight to some extent on a part of it, The drone can still maintain its balance to compensate for the imbalance under the other arms as much as the weight of the smartphone.After that lead mental was attached.The smartphone was fixed on the drone arm in the middle of the arm in a protective frame with tape adhesive tightly and the errand the aerodynamic.The energy-stored circuit was installed on drone batteries, so the electricity is taken from the nanogenerators and stored itself.The vibration sampling rate frequency was 500 Hz.The lifting power of rotor drones is provided by the rotation of the propeller and its engine.The brushless motor in the hexarotor drones is installed at the end of the arms.There is more vibration where the drone motors are installed that transmit vibration over the drunk arm.In this study, experiments and data collection were repeated 30 times.Android application saved each vibration data in 2 ms.Therefore, during 1 min of flight in hover mode, 30,000 vibration data in the time domain were saved.The duration of data collection varied between 1 and 3 min.
Different parts of a working UAV are subject to vibration.Because the motors are mounted on the arms, the arms vibrate and change shape.When the drone's engines turned, different parts of the drone generated vibration, and the vibration produced electricity in the nanogenerator.The harvester converts the vibration into electrical energy.Then, the voltage was transferred to the circuit through the electrodes and stored in the capacitor.For 1 min, after the hexa rotor landing and turned off, the voltages were measured by a multimeter and recorded and stored in the computer.To harvest electrical energy from the nanogenerator, a circuit consisting of a capacitor with a capacity of 47 μF, a 1 kΩ resistor, and an LED lamp was used.Also, in order to display the output voltage, a multi-meter device was used by connecting two nanogenerator electrodes.

Signal processing and calculation of statistical features
The amplitude and distribution of vibration signals in the time domain were different in different vibration conditions of the UAV arm.In signal processing engineering, it is one of the most important and widely used topics.Any alternating function can be expressed as a set of sine and cosine functions called a Fourier series.A Fast Fourier Transform (FFT) decomposes a string of values into components with different frequencies.FFT is a way to calculate the same results more quickly [30,31].The function f(x) is a periodic function, and its integral in the period is limited, and the number of minimum and maximum and discontinuities is limited.In that case, it can be written with a series of basic functions (sine and cosine).Energy harvesting and ANFIS modeling  5 It converts analog signals x(t) in the time domain into digital data in the frequency domain.It also receives the analog signals output from the sensors and calculates the spectral coefficients of these signals (a 0 , a n , and b (n) ) in terms of frequency using the above equations.From the calculation of N samples x 1 (t i ) and x 2 (t i ) and … and x N (t i ), the discrete form of the Fourier transform is obtained: 1, 2, 3,…, .
In the above relationship, the digital spectral coefficients a 0 , a i , b(i) according to the above equations: where N is equal to 2 to the symbol of an integer and for each analyzer, and N is an eigenvalue.Since the maximum displacement and strain occurs at the resonant frequency, the maximum voltage is also produced in this state [32].For example, the time and frequency spectrum of a data set with 500 UAV vibration signal points is shown in Figure 7.
The extraction of best features is important in the analysis of vibration signals.In the feature extraction phase, the most critical parameters affecting the vibration of the UAV arm were revealed.In the pre-processing UAV vibration signal stage, due to the large number of vibration waves of the UAV, in this study, the vibration data were considered only in the hover state of the UAV.For this purpose, the number of data selected in 30 s (from 10,000 to 40,000 ms) was considered.The data related to time losses and multi-rotor take-off and landing were analyzed and modeled in the process.The FFT frequency domain data points, which were 1,200 in total, were used to determine the feature vector.Therefore, the most important features were calculated using some feature parameters from time and frequency domains.The number of 10 statistical characteristics of the vibration signals was calculated and their relationships are given below.
Maximum Value: The maximum value of a vibration data set.
Minimum Value: The minimum value of a vibration data set.
Sum: The total sum of the values of a certain sample of vibration signals.
Mean: In a vibration data set, the sum of all data points divided by the number of data points is mean.
Skewness: If the tension is zero, the vibration data distribution is normal, but if the tension is negative, the data distribution curve is lying at the peak.This asymmetry of the normal distribution is expressed in a set of vibrational data describing Skewness.
( ) )( ) Range: The difference between the highest and lowest vibration data values in a data set.
Variance: S² is a dispersion index that is the mean of the sum of the squares of the deviation of the score from the mean.Here, x i is a signal point and i represents the number of data points.

ANFIS modeling
MATLAB version 7.10.0.499 software was introduced to the Takagi method of ANFIS to model three sets of training and test fuzzy input-output patterns.Table 1 in this study shows the results of the two ANFIS models for vibrational energy harvesting from the nanogenerator mounted on the drone.
Each feature included 90 vibrational data that for ANFIS modeling was used.The input membership function (MF) in the A and B ANFIS modeling and processes was the Gaussian and glimf membership, respectively.The output MF was linear.The learning hybrid method was used.About 70% of the vibration data were for training.After training, 30% of vibration signals were considered for network testing.The total number of input data of ANFIS in each model was 360 vibration data.The number of training and testing vibration data was 189 and 81, respectively.The ANFIS inputs included the selected vibration features and the outputs included the stored voltage.The number of network training MF in the input structure of the network was equal to 13, and their type was selected with the lowest amount of error, and the relationship between the training error of the network and number of epochs in ANFIS models was obtained.FIS structure and shape of the MF modeling nanogenerator energy harvesting on the drone were obtained.

Results and discussion
In this study, a piezoelectric nanogenerator was built and installed on the arm of a hexacopter agricultural sprayer.The vibrations of the UAV led to the deformation of the nanopizoelectric, and electrical energy was produced.The vibration signals were recorded during the specified period of time and the test process, and the amount of electrical energy produced was also recorded at each stage of the tests.According to the waveform, 30,264 vibration signal points in a period of 60,646 ms are recorded.The UAV acceleration is shown in flight hover mode in three directions x, y, and z.It can be concluded from the shape of the graph that the range is less at the beginning and end of the graph (Figure 8).This matter is related to the fact that the data collection system is active in drone mode and shows almost the vibrations of the environment.However, a most significant increase in the vibration amplitude can be seen on both sides of the graph.This is due to the take-off and landing of the bird during the test.In the middle of the graph, a uniformly symmetrical shape is observed, which corresponds to the more stable flight mode and the hover mode.The results show that the highest range of vibrations occurs during the UAV take-off and landing.According to Figure 9, the maximum range of acceleration in the x, y, and z directions is 20.6, 6.09, and 13.7 M/s 2 in frequencies 54.3, 49.8, and 0.49 Hz, respectively, and the maximum range of acceleration occurred in the direction of the x-axis of the UAV arm.A screenshot of the Idynamic application of the smartphone during the vibration data transfer of the drone arm is shown in Figure 9.
Figure 9 shows the position of the UAV arm moving in vibration conditions during the test and flight stages for about 1 min (0-66.9S).According to the graph's figure, at the beginning of the graph, within 3 s of starting the data collection, the graph has displacement, which is the amount of displacement related to the human intervention of the activation, including the application, and placing the drone in the safe flight position.After that, the displacement graph is stabilized.When the UAV is turned on and the vibrations start, the displacement in three directions, x, y, and z, increases.However, two peaks are observed in the displacement wave.The first peak is displayed with a displacement of 0.42 m along the x-axis, 0.23 m along the y-axis, and 0.26 along the z-axis.This amount of displacement is related to the take-off time of the drone.After that, the pilot put the UAV in hover mode and the displacement waveform became more uniform.Another peak of the displacement occurred at 50 s, which is related to the landing time of the UAV, and after that, the UAV was turned off in the time range of 60-63 s, and the displacement in all three directions was close to zero.Finally, the data collection process was stopped, and the vibration data was recorded.In order to calculate the effect of vibration on the piezoelectric performance of the nanogenerator as well as the deformation effect, the generated electric voltage was measured in each step of the test for 1 min.An electrical circuit design was used.The circuit made by the capacitor stored the value of the output voltage in the nanogenerator.The mechanical vibrations created by the UAV arm caused deformation and strain in the active materials of the nanogenerator, which led to the generation of voltage.The quality, material, structure of the nanogenerator, the applied forces, the size of the nanogenerator, and the duration of time when the nanogenerator was subjected to vibration were effective on the produced voltage.In order to find the resonance frequency, the experiments were repeated several times.Since the instantaneous energy produced by a piezoelectric harvesting machine is not enough to power the various electronic parts of the UAV, a storage medium such as a battery or capacitor is often used to store the harvested energy before use temporarily.Several 30 samples of nanogenerators were tested and repeated three times for each sample.The vibration data collection was repeated 30 times, and the stored voltage value was recorded each time (Figure 10).
In the graph of Figure 10, the output voltage of the nanogenerator is shown at each time of data collection.As can be seen from the graph, the highest recorded voltage was 64 mV during the drone flight time at 85.5 min in test No.26.Based on the results of each flight test, there are different output voltages due to the different times taken for safe landing and take-off of the UAV.There are different output voltages, and the results showed that usually, with the increase of the flight time, more voltage is stored in the circuit, which shows that the nanogenerator has been exposed to vibrations for longer.
The results of a number of researches showed that the development of nanostructures of lightweight energy  harvesting nanogenerators, along with soft computing techniques, can lead to an optimized piezoelectric vibration-based energy harvester and improve its performance by increasing the effective frequency bandwidth and increasing the amount of voltage that can be taken [18].
Through modeling, analysis, optimization, and design of the energy harvester system for RFID hybrid systems, various tests were performed on the voltage harvester experimental setup.The vibration-based piezoelectric energy harvester was able to generate up to 4.7 mW of electricity to Energy harvesting and ANFIS modeling  9 be used to power ultra-low-power RFID components.The correctness of the designed fuzzy inference system was also confirmed using the numerical-analytical method [18].In order to experimentally analyze energy harvesting using PZT materials under the dynamics, loading with three boundary conditions and checking with two circuits were successfully developed.The number of neural network inputs included three variable dynamic load parameters and the number of beats and piezoelectric mode (simple, fixed, and circular).Two circuits were tested 120 times.Artificial neural networks were used to model training between three input parameters and one output parameter.In addition, the genetic algorithm was used on the artificial neural network model.The goal was to maximize the output of the system with the input parameters.Based on the results, in the number of loads of 26 blows and with 44 N of force, with circular support connected to the piezoelectric, the maximum voltage was twice as high as 16.9 V.After that, the input parameters were optimized to maximize the output voltage by GA and ANN [26].Also, piezoelectric smart arms were able to provide up to 5.35 mW during a steady flight, enough voltage to power low-power sensors used to monitor UAV applications [21].In this study, vibration analysis was not done based on the time-amplitude diagram, and frequency analysis was omitted.11 showed.So, the average R and MSE of the model were equal to 0.84 and 0.49, respectively.Then, the graph of the levels and shape of the fuzzy rules was obtained in the modeling process with ANFIS in each state.About nine fuzzy rules were obtained in ANFIS modeling.Each signal point can be seen clearly (Table 2).In Figure 11, the output of the model predicts an ANFIS against the prediction value and detection error.There are 30 output data, and 30 vibration signal responses were identified.For example, in the right shape of the ANFIS model A, the data number of voltage is produced by UAV vibration.In the graph of Figure 11, the output voltage of the response experimentally compared to the voltage predicted by the model is shown to evaluate the model of this study.Based on the results of ANFIS modeling compared to the experimental method, it was confirmed that the error rate was low.The lower the MSE value and the lower the error rate of the predicted values compared to the training values, the more accurate the designed model is and the better it can help us.In Figure 12, the two-dimensional and three-dimensional diagrams show the changes in input parameters' influence level, which includes time data and feature data of maximum vibration signals in Model A. Based on the results shown above on the right side, there is a good correlation between time input and voltage output.This figure indicates that the voltage produced by the nanogenerator increases with the increase in the drone's flight time.Also, in the diagram on the left side, it is shown that the level of influence of the maximum extracted feature is that the input value varies with the output.As Figure 12 (right-below) shows, the three-dimensional pattern in the form of color dispersion shows the levels of influence of the input parameters on the amount of voltage produced by the nanogenerator installed on the drone, which can be used to analyze and optimize the performance of the nanogenerator.In the graph, the experimental values are shown in x-axis against the values predicted by the ANFIS model, and the experimental values are somewhat close to the predicted values.Based on the results of the ANFIS model, the output can help us in optimizing the performance of the parameters affecting the voltage produced by the nanogenerator installed on the UAV.Also, the ANFIS model can be used as an effective tool in predicting the amount of electrical energy produced according to the vibration state of the UAV in the future.In this study, there was a difference between the time and frequency vibration spectra in different states of the UAV arm.In agricultural drones, depending on the mass of the drone and the mass of the solution inside its tank, the flight time of a 20-kg drone with a 12,000-mA h battery is approximately 10 min.During frequent take-offs and landings or sudden changes in the rotors, the battery consumption increases and the flight time decreases.Installation of piezoelectric nanogenerators based on vibration can produce a small amount of energy.The electric energy that can be collected through the vibrations of the drone depends on the vibration behavior, the amplitude of the vibrations, and the piezoelectric structure used, and it is complicated to determine the exact amount of the production voltage, for this reason, although we were able to produce some electricity energy, this amount is not enough to increase the flight time of the drone, definitely.Finally, the results showed that vibration signals and piezoelectric nanogenerators can also be used in the UAV status monitoring program.

Conclusion
In this study, the amount of electrical energy produced by the manufactured nanogenerator piezoelectric, although it could not produce enough electrical energy required for the flight time of the UAV, was able to be used by producing a small amount of electrical energy to meet the needs of the sensors and monitoring programs of the UAV.The maximum voltage recorded during the 30-s flight time of the tested UAV was 64 mV at 3.54, 49.8, and 0.48 Hz along the x,  y, and z axes, respectively, based on the results, ANFIS modeling for predicting the energy harvesting of nanogenerator was effective.The results indicated that loaded vibration data had a significant relationship with the model's performance and the predicted data.The results of feature selection by trial and error method showed that among the selected features, feature time and maximum and range vibration signals were used as the best features in the modeling process.The analysis of the time domain of the signals was complex and showed the shape of the frequency graphs in each of the different vibration modes.
According to the output table of the best ANFIS model for modeling model 1, the highest R-value was with an accuracy of 0.85.Based on the results, it is also possible to improve the electrical energy that can be produced with the help of nanostructure used in piezoelectric and proper use in drones.Vibrations in UAVs, especially rotary UAVs with agricultural applications, can be a good source of electricity generation using piezoelectric materials.It is necessary to use this source well in order to produce more electrical energy.In this regard, it is suggested to carry out more research on different locations and the use of more piezoelectric materials in the structure of multi-rotors and increasing the flexibility and power generation with the help of nanostructures in piezoelectric materials and the ability to store more electricity, as well as other methods of optimizing energy harvesting.And it is necessary to conduct more studies to increase the piezoelectric properties of the materials used with the help of nanostructures and how to store energy with the help of nanomaterials.UAVs as well as UAV status monitoring programs should be used well.

Figure 1 :
Figure 1: Chart form of research steps.

Figure 7 :
Figure 7: Example of vibration spectrum in time and frequency domain of UAV vibration data.

Figure 8 :
Figure 8: The acceleration of UAV arm under test in hovering flight mode.

Figure 9 :
Figure 9: Vibration spectrum of UAV arm during data collection and the example of displacement of drone arm in one of the data collections.

Figure 10 :
Figure 10: Output voltage results of the nanogenerator mounted on the drone arm.

Figure 12 :
Figure 12: The effect surface of UAV vibration on nanogenerator voltage and the accuracy of the performance of neuro-fuzzy inference systems in nanogenerator in various research studies is acceptable.

Table 1 :
ANFIS modeling specifications The use of polymer nanofibers in the energy supply of drones and the vibration modeling of piezoelectric nanogenerators with ANFIS that have high accuracy were among the superior aspects of innovation of this research.Examples of the vibration spectrum of the UAV in the hover state are shown in Figures8 and 9.As can be seen, the characteristics were different in the rounds.In models, the best features consist of maximum and time in model A, and range and time in model two were selected among the ten desired features maximum, minimum, range, mean, variance, standard deviation, and kurtosis.Of course, suitable data mining techniques can extract better features of vibration signals from UAVs.In order to further reduce the number of features and choose the most suitable features for modeling, considering that in both rounds of model A and model B, the number of common features was high, so among the 10 features, the best features were selected for modeling.Two features, range at model A and minimum at model B, were different in both rounds and were not common, so they were removed from the modeling process to reduce the number further.Statistical indicators such as mean squared error (MSE), mean absolute error (MAE), and sum squared error (SSE) were used to examine model performance.The average value of correlation coefficient (R) in the models of one cycle of model A was 0.85 and for the model of model B, it was 0.84.The results show more validity of modeling.The predicted values were close to the output values, and the error rate was negligible, which was a sign of the high efficiency of the introduced ANFIS modeling system.The high correlation coefficient value indicated a strong correlation between the experimental values, and those values predicted by the ANFIS modeling system.The value of the R the relationship between it shows output value and input values.Accordingly, if the values are close to one, it indicates that there is a close relationship between the values taught and the values predicted by the ANFIS network, and if R is close to zero, it indicates a random relationship.Based on the available results, acceptable accuracy has been obtained in the rest of the models.Based on the results, the small values of SSE, MAE, and MSE indices as well as the R index values and close to one in each round indicate the acceptable accuracy of the ANFIS modeling.After building the network and testing its error percentage, the network was saved to obtain the final result.The error graph was also obtained according to the training courses.The results of the experimental values were obtained for the predicted values.The amount of insignificant errors regarding each of the input signals for the output signals of the graphs in Figure

Table 2 :
The output of the best ANFIS model for modeling