Research on Flow Field Perception Based on Artificial Lateral ... - MDPI

sensors Article

Research on Flow Field Perception Based on Artificial Lateral Line Sensor System Guijie Liu 1, * ID , Mengmeng Wang 1 ID , Anyi Wang 1 , Shirui Wang 1 , Tingting Yang 1 , Reza Malekian 2, * ID and Zhixiong Li 3 ID 1

2 3

*

Department of Mechanical and Electrical Engineering & Key Laboratory of Ocean Engineering of Shang Dong Province, Ocean University of China, Qingdao 266100, China; [email protected] (M.W.); [email protected] (A.W.); [email protected] (S.W.); [email protected] (T.Y.) Department of Electrical, Electronic & Computer Engineering, University of Pretoria, Pretoria 0002, South Africa; [email protected] School of Mechanical, Materials, Mechatronic and Biomedical Engineering, University of Wollongong, Wollongong 2522, NSW, Australia; [email protected] Correspondence: [email protected] (G.L.); [email protected] (R.M.); Tel.: +86-532-6678-1021 (G.L.); +27-124-2043-05 (R.M.)

Received: 7 February 2018; Accepted: 3 March 2018; Published: 11 March 2018

Abstract: In nature, the lateral line of fish is a peculiar and important organ for sensing the surrounding hydrodynamic environment, preying, escaping from predators and schooling. In this paper, by imitating the mechanism of fish lateral canal neuromasts, we developed an artificial lateral line system composed of micro-pressure sensors. Through hydrodynamic simulations, an optimized sensor structure was obtained and the pressure distribution models of the lateral surface were established in uniform flow and turbulent flow. Carrying out the corresponding underwater experiment, the validity of the numerical simulation method is verified by the comparison between the experimental data and the simulation results. In addition, a variety of effective research methods are proposed and validated for the flow velocity estimation and attitude perception in turbulent flow, respectively and the shape recognition of obstacles is realized by the neural network algorithm. Keywords: artificial lateral line system; hydrodynamic simulation; flow field perception; velocity estimation; neural network

1. Introduction Most fishes are dependent on the lateral organs to detect their surrounding environment, track moving targets and avoid obstacles [1]. In steady flow, they rely on lateral sensing information to estimate flow velocity [2], direction [3], realize rheotaxis and reduce energy consumption in different swimming state [4]. So, the function of the fish lateral-line provides a new idea for underwater perception. At present, many researchers have carried out detailed studies about artificial lateral-line sensor’s design, fabrication and optimization [5]. They have developed different sensing systems, conducted parametric underwater experiments and established related algorithms [6]. Pandya produced an artificial lateral-line system equipped with commercial hot-wire anemometer sensors and micro-mechanical sensors, which successfully located a moving dipole source [7]. Venturelli developed a rigid parallel distributed lateral line system installed with 20 pressure sensors, which achieved the purpose of fluid environment identification and Karman vortex streets detection [8]. Chambers produced a three-dimensional fish head with 33 pressure sensors (MS5401-AM) to study the fluid interaction [9]. Levi DeVries designed an artificial system based on 8 IPMC sensors and 4 pressure sensors, which estimates flow parameters for feedback control [10]. The experimental carrier shapes of

Sensors 2018, 18, 838; doi:10.3390/s18030838

www.mdpi.com/journal/sensors

Sensors2018, 2018,18, 18,838 x FOR PEER REVIEW Sensors

2 2ofof2424

pressure sensors, which estimates flow parameters for feedback control [10]. The experimental above researches focused on mainly bionic fishes buton there arefishes few studies directly on applications of carrier shapes ofmainly above researches focused bionic but there are few studies directly underwater vehicles. on applications of underwater vehicles. To lateral-line system systemin inunderwater underwaterequipment equipmentsuch suchas Topromote promotethe theapplication application of of the the artificial lateral-line asunderwater underwater vehicles, a cylindrical carrier was utilized a main research object in this paper. vehicles, a cylindrical carrier was utilized as aasmain research object in this paper. The The lateral line sensing mechanism was studied in the perspective of biomechanics, which laid lateral line sensing mechanism was studied in the perspective of biomechanics, which laid a biological afoundation biological foundation for the design of artificial lateral-line system. The optimal sensor distribution for the design of artificial lateral-line system. The optimal sensor distribution model and model and underwater environment perception algorithm were obtained through the hydrodynamic underwater environment perception algorithm were obtained through the hydrodynamic simulation. simulation. Anlateral artificial lateral linebased systemonbased on the optimal topology the sensors designed. An artificial line system the optimal topology of theofsensors was was designed. The The underwater experiments verified validity of numerical simulation a comparison between underwater experiments verified thethe validity of numerical simulation by by a comparison between the the experimental data and the simulation results. The effective methods of flow velocity estimation, experimental data and the simulation results. The effective methods flow velocity estimation, attitude attitudeperception perceptionand andobstacle obstacleidentification identificationbased basedon onthe theartificial artificialsystem systemwere wereput putforward forwardfrom from the theverification verificationexperiments. experiments. 2.2.Biomechanical BiomechanicalModel ModelofofLateral LateralLine Line Lateral Lateralline lineisisaahydrodynamic hydrodynamicreceptor receptorsystem systemininfishes fishesand andamphibians, amphibians,which whichmainly mainlysenses senses the local water pressure, vibration and other hydrodynamic stimuli [11]. The organizations the local water pressure, vibration and other hydrodynamic stimuli [11]. The organizationsofoflateral lateral organs organsare arevarious variousinindifferent differentspecies. species. Biologists Biologists found found that thatthe thelateral lateralorgan organwas wascomposed composedofof neuromasts, neuromasts,which whichare aredistributed distributedon onboth bothsides sidesof ofthe therunning runningfish fishbody. body.According Accordingtotothe thelocation, location, neuromasts are divided into two types, superficial neuromasts (SNs) and canal neuromasts neuromasts are divided into two types, superficial neuromasts (SNs) and canal neuromasts(CNs) (CNs)[12]. [12]. Their biomechanical model is shown in Figure 1. Their biomechanical model is shown in Figure 1.

Canal Neuromast

Superficial Neuromast Flow

High pressure

Canal Flow

Low pressure Local Flow

Cupula Hair cell

Figure 1. Schematic of Superficial Neuromasts and Canal Neuromasts. Figure 1. Schematic of Superficial Neuromasts and Canal Neuromasts.

Netten et al. established a biomechanical model of lateral sensing neuromasts, which provided et the al. established a biomechanical model ofsystems lateral sensing neuromasts, which provided the Netten basis for development of artificial lateral line [13]. The quantitative analysis of his the basis for the development of artificial lateral line systems [13]. The quantitative analysis frequency response formula facilitated the understanding of the hydrodynamic input and ofperceived his frequency response formula facilitated the understanding of the hydrodynamic input and output. perceived The output. frequency response of SNs is defined as [14] The frequency response of SNs is defined as [14] (

)

( )   q ( )= =− 1− ( − H (δ1+i) ) + 3 2π f ibm υ( H ) 2ibw  2 iπ+f bm δ4 (i j H 4  EI ) SSN ( f ) = =− 1− + ∑ Cj 4 U∞ 2π f bm 2EI + iπ f bm δ j =0 is free-stream velocity, where ( ) is cupular deflection at the height of the beam tip, where υ( H ) is cupular deflection at the height of the beam tip, U∞ is free-stream velocity, = −4 − ( + ) + iπ 2 µk bm = −4πµk − iπ (ρw + ρm ) a2 ω + L = −4 −2 + iπ 2 µk 2 where μ is the dynamic viscosity k is−viscous bwof=fluid. −4πµk 2iπρw aforce ω +coefficient, L is radius of the hemisphere, C j is a length of 4 constant sequence,

density, H is the height of the cartilage peak L is defined as:

w m

(1) (1)

(2) (2) (3)

is the density of water, (3)a is cartilage peak matrix

Sensors 2018, 18, 838

3 of 24

where µ is the dynamic viscosity of fluid. k is viscous force coefficient, ρw is the density of water, a is radius of the hemisphere, Cj is a length of 4 constant sequence, ρm is cartilage peak matrix density, H is the height of the cartilage peak L is defined as: a L = γ + ln( √ ) 2δ

(4)

where γ is Euler’s constant (γ ≈ 0.5772), δ is boundary layer thickness. We focused on the bionics analysis rather than mathematics. SNs are similar to displacement transducers and sensitive to flow velocity because the displacement of the cupula is relative to viscous drag and consequently to the velocity of water particles. So SNs can estimate the flow velocity and orientation on their body surface. The frequency-dependent sensitivity of CN cupula can be defined as [15]: √

SCN ( f ) =

Y0 ( f ) 1 = V0 ( f ) 2π f t

Nr + i

 1

f 2 2 + 13 i fft 2 (1 + i ) f t  3  2 √ f f 2 2 − 13 fft f t − 2 (1 − i ) f t

1+

(5)

where Y0 ( f ) is vibration amplitude of cartilage peak amplitude, V0 ( f ) is external flow velocity amplitude, f t is defined as µ (6) ft = 2πρw a2 where µ is the dynamic viscosity of fluid. ρw is the density of water, a is radius of the hemisphere, f / f t is Stokes number, Nr is defined as the resonance factor that governs the resonance properties of a cupula. CNs are regard as acceleration sensors because the canal fluid velocity is practically proportional to the first full derivative of the flow speed outside the canal [16]. And CNs can also be regard as pressure gradient sensors because the water acceleration is proportional to the pressure gradients. Thus, CNs detect the pressure distributions in their lateral line canals [15]. SNs can distinguish fields in spatially uniform flow and turbulent flow while CNs only respond non-uniform flow fields, such as water fluctuations generated by a vibrating sphere or a swimming fish [17]. The artificial lateral line system combined features of SNs and CNs. Miniature pressure sensors are arranged on the surface of the carrier and the flow field characteristics are obtained by the surrounding pressure field. There are three main research goals: 1. 2. 3.

Flow velocity estimation Attitude perception Obstacle identification.

All the experiments are carried out in uniform flow or regular non-uniform flow (Karman vortex streets). 3. Optimal Topology of Sensors The artificial lateral line system is equipped with a number of pressure sensors, so the distribution of sensors should be studied. More sensors should be placed on the most sensitive locations [18] of the water flow environment. Imitating an underwater aircraft shape [19,20], the artificial carrier is cylinder with a hemispherical head. Simulation is carried out through ANSYS FLUENT with basic parameters in the Table 1.

Sensors 2018, 18, 838

4 of 24

Table 1. The basic parameters of simulation.

Sensors 2018, 18, x FOR PEER REVIEW

4 of 24

Sensors 2018, 18, x FOR PEER REVIEW

4 of 24

Mesh (Icem) Table 1. The basic parameters of simulation. Fluid Dimensions 1 mTable × 3 m1. The basic Carrier Dimensions 0.1 m × 0.4 m parameters of simulation. Mesh Grid (Icem) Number of Grids 232,776 type Unstructured grids Fluid Mesh (Icem) 1 m × Hydrodynamic 3m Carrier Dimensions 0.1 m × 0.4 m Simulation (Fluent) Dimensions Fluid 1m×3m Carrier Dimensions 0.1 m × 0.4 m Standard Number of Grids 232,776 K-ε Grid type Unstructured grids Dimensions Boundary conditions Velocity inlet/pressure outlet Physical model model Hydrodynamic Simulation Number of Grids 232,776 Grid type(Fluent) Unstructured grids Inlet velocity −1K-m/s Reynolds number 49,900–499,000 Simulation (Fluent) Hydrodynamic Standard model Physical model Boundary conditions Velocity inlet/pressure outlet  model Standard Physical model Boundary Velocity 49,900–499,000 inlet/pressure outlet Inlet velocity −1 Km/s Reynoldsconditions number Inlet velocity −1 m/s Reynolds number 49,900–499,000

In the three-dimensional simulation analysis, it can conclude that the static and dynamic pressure In the three-dimensional simulation analysis, it can conclude that the static and dynamic value is almost the same in the cross-section perpendicular the carrier axis, sostatic it canand be simplified as In the three-dimensional simulation analysis, it perpendicular cantoconclude that pressure value is almost the same in the cross-section to thethe carrier axis, sodynamic it can be two-dimensional flow fieldthe simulation. The plane mesh result is Figure As shown in Figure pressure value is almost same the cross-section perpendicular the2a. carrier axis, can bein2b, simplified as two-dimensional flowin field simulation. The plane mesh to result is Figure 2a.so Asitshown the pressure data of the surface of the carrier is extracted along the edge of the section. The zero point simplified as pressure two-dimensional flow field simulation. Theisplane meshalong resultthe is Figure Assection. shown The in Figure 2b, the data of the surface of the carrier extracted edge of2a.the is theFigure front 2b, endthe ofpressure the carrier, the upper part is the negative direction and the lower part is the positive data surface ofupper the carrier along the edgeand of the The is zero point is the front end of of thethe carrier, the part is is extracted the negative direction thesection. lower part zero point is the front end of the carrier, the upper part is the negative direction and the lower part direction. Starting from zero, the length of the curve along the edge of the section is defined as “carrier the positive direction. Starting from zero, the length of the curve along the edge of the sectionis is the length.” positive direction. Starting from zero, the length of the curve along the edge of the section is surface defined as “carrier surface length.” defined as “carrier surface length.”

(a) (a)

(b) (b)

thegreen green area Figure 2.2.The results of the meshing the definition definition of of the theaxes. axes.(a) (a)The Theplane plane mesh;the area Figure The resultsof ofthe themeshing meshing and and Figure 2. The results andthe the definition of the axes. (a) The mesh; plane mesh; the green represents the fluid domain and the oval shape represents representsthe thecarrier. carrier.(b) (b)The Thedefinition definitionofofaxes; axes;the the represents the fluid domain and the oval shape area represents the fluid domain and the oval shape represents the carrier. (b) The definition of axes; dark blue oval shape represents the carrier in the the picture. picture. dark blue oval shape represents the carrier in the dark blue oval shape represents the carrier in the picture.

Simulation of 0.1 0.1 to to 11 m/s m/s calculated calculatedby byFLUENT FLUENTare areintegrated integratedbyby Simulationresults results with with flow flow velocity velocity of Simulation flow3. MATLAB and shown in MATLAB andresults shownwith inFigure Figure 3.velocity of 0.1 to 1 m/s calculated by FLUENT are integrated by MATLAB and shown in Figure 3.

(a) (a)

(b) (b)

Figure3.3.Pressure Pressure distribution distribution of of carrier carrier surface at velocity. (a) Dynamic pressure; Figure surface at at different different flow flow Figure 3. Pressure distribution of carrier surface different flow velocity. velocity. (a) (a)Dynamic Dynamicpressure; pressure; (b) Static pressure. Static pressure. (b) (b) Static pressure.

As shown in Figure 3 is the relationship between surface pressure, velocity and geometry, that As shown in Figure 3 is the relationship between surface pressure, velocity and geometry, that is, the overall pressure distribution model, the specific data can be obtained by picking up surface Figure 3distribution is the relationship surface pressure, velocity geometry, that is, is,As theshown overallinpressure model, between the specific data can be obtained byand picking up surface points. The location of extreme points is shown in Table 2. As the flow velocity increases, the value thepoints. overallThe pressure distribution model, the specific data can be obtained by picking up surface points. location of extreme points is shown in Table 2. As the flow velocity increases, the value of surfaceofpressure gradually increases and the2.position the extreme increases, point doesthe notvalue change. Theoflocation extreme points isincreases shown inand Table As the of flow of The surface surface pressure gradually the position of thevelocity extreme point does not change. The pressure slope of maximum point is the highest, thus it is the most sensitive to flow velocity and pressure gradually andpoint the position of the extreme does sensitive not change. The pressure slope pressure slope ofincreases maximum is the highest, thus it ispoint the most to flow velocity and suitable for pressure sensors placement. By observing the characteristics of pressure changes, we can suitable for pressure sensors placement. By observing the characteristics of pressure changes, we can

Sensors 2018, 18, 838

5 of 24

Sensors 2018, 18, x FOR PEER REVIEW

5 of 24 of maximum point is the highest, thus it is the most sensitive to flow velocity and suitable for pressure sensors placement. By observing the characteristics of pressure changes, we can see that sensors are see that sensors are deployed at the mid-point of the pressure distribution and at the maximum deployed at the mid-point of the pressure distribution and at the maximum point and the minimum point and the minimum point of the pressure data, which can detect the pressure changes more point of the pressure data, which can detect the pressure changes more sensitively, so that the position sensitively, so that the position is suitable for arranging the sensor. In addition, the pressure change is suitable for arranging the sensor. In addition, the pressure change at the zero point of the static at the zero point of the static pressure is 0, which can be used as the pressure reference point. pressure is 0, which can be used as the pressure reference point. Therefore, the pressure sensor is also Therefore, the pressure sensor is also suitable here. suitable here.

Table 2. The of extreme extreme points. points. Table 2. The location location of Extreme Points Extreme Points

Dynamic Pressure Dynamic Pressure −0.387 −0.066−0.387 Maximum point coordinates 0.057 −0.066 Maximum point coordinates 0.384 0.057 0.384 −0.456 Minimum point 0 −0.456 coordinates Minimum point coordinates 0 0.456 0.456

Static Pressure Static Pressure −0.456 0 −0.456 0 0.456 0.456 −0.387 − 0.387 −0.066 −0.057 0.066 0.057 0.384 0.384

According to the variation of static pressure between different flow rates, it can be seen that the to thezero variation rates, itinlet canand be seen that 0.03 pressure , because between the initialdifferent pressureflow of velocity pressure staticAccording pressure keeps near Xofstatic the static pressure keeps zero nearconditions, X = ±0.03, velocity inlet and outlet is set to 0 under simulation so itbecause can be the usedinitial as thepressure basis forofjudging the current pressure outlet is set to 0 under simulation conditions, so it can be used as the basis for judging depth of carrier. Due to the maximum point of hydrodynamic data and the minimum point of the current depth of carrier. Due to the point position of hydrodynamic and the minimum point hydrostatic data are rather special andmaximum the coordinate needs to data be determined according to of hydrostatic data are rather special and the coordinate position needs to be determined according to the calculation. Because of the symmetrical distribution on the left and right, first take the average of the calculation. Because of the symmetrical distribution on the left and right, first take the average of the abscissa as follows: the abscissa as follows: x1  (0.075  0.066  0.057  0.066) 4  0.066 m (7) x1 = (0.075 + 0.066 + 0.057 + 0.066)/4 = 0.066 m (7) 1  x1 r  0.066 0.05  1.32 rad  75.67 (8) θ1 = x1 /r = 0.066/0.05 = 1.32 rad = 75.67◦ (8) x2  [(0.914 / 2  0.387)  (0.914 / 2  0.384)] 2  0.0715 m (9) x2 = [(0.914/2 − 0.387) + (0.914/2 − 0.384)]/2 = 0.0715 m (9)    x2 r  0.0715 0.05  1.43 rad  81.97 ◦ (10) = 0.0715/0.05 = 1.43 rad = 81.97 (10) θ2 =2 x2 /r

0.03+ + 0.03 2 =0.03 mm x3 x=3 (（ 0.03 0.03） 0.03 )/2

(11) (11)

 ◦ =0.6 0.6rad rad = 34.4 θ3 =3x x3 = r 0.03/0.05 0.03 0.05 =  34.4 3 /r

(12) (12)

According to the geometric relationship between the surface length and diameters of hemispherical head, the position of the feature point can be obtained, obtained, shown shown in in Figure Figure 4. 4.

O

75° 30° -30°

82°

x

-82°

-75°

y Figure 4. 4. The The pressure pressure sensitive sensitive point point on on carrier carrier surface. surface. Figure

When the carrier and the flow direction are 15°, 30°, 45° respectively and flow velocity is fixed at 0.5 m/s, the surface pressure distribution is shown in Figure 5. Due to the vortex at this time and the surface pressure is related to vortex shedding frequency, the simulation time is fixed at 2.5 s.

Sensors 2018, 18, 838

6 of 24

When the carrier and the flow direction are 15◦ , 30◦ , 45◦ respectively and flow velocity is fixed at 0.5 m/s, the surface pressure distribution is shown in Figure 5. Due to the vortex at this time6 and Sensors 2018, 18, x FOR PEER REVIEW of 24 the surface pressure is related to vortex shedding frequency, the simulation time is fixed at 2.5 s. Figures the the cylindrical area pressure is almostisunchanged in symmetry Figures indicate indicatethat that cylindrical area pressure almost unchanged indistribution symmetry ◦ ). As the angle increases, the symmetry is broken, the dynamic when the carrier is facing the flow (0 distribution when the carrier is facing the flow (0°). As the angle increases, the symmetry is broken, pressure of incident surface flow (y < surface 0) increases the countercurrent surface pressure the dynamic pressureflow of incident (y < 0) increases the countercurrent surfacedecreases, pressure the stagnation point moves, the original two maximum points reduced to one. decreases, the stagnation point moves, the original two maximum points reduced to one.

(a)

(b)

(c)

Figure 5. The surface pressure curve in different angles under 0.5 m/s. (a) A schematic of angle; Figure 5. The surface pressure curve in different angles under 0.5 m/s. (a) A schematic of angle; (b) Dynamic pressure; (c) Static pressure. (b) Dynamic pressure; (c) Static pressure.

The arrangement of the sensor spacing in the cylinder area can be calculated from pressure of the sensorand spacing in the cylinder area can be calculated fromthat pressure slope slopeThe of arrangement incoming flow surface the sensitivity of pressure sensor. Assume the sensor of incoming flow surface and the sensitivity of pressure sensor. Assume that the sensor sensitivity is sensitivity is Sp, the sensor spacing is x and the pressure slope is k p . x can be expressed as: Sp , the sensor spacing is x and the pressure slope is k p . x can be expressed as:

 xx =

sspp kkpp

(13) (13)

The pressure system cancan capture are are proportional to the spacing, so there pressuregradients gradientsthat thatthe the system capture proportional tosensor the sensor spacing, so is a minimum distance between the two at that the pressure difference between two there is a minimum distance between thesensors; two sensors; at time that time the pressure difference between ◦ , simulation sensors is exactly the sensor resolution. AssumeAssume that the that sensitivity to the angle two sensors is exactly the sensor resolution. the sensitivity to is the15angle is 15°, showing theshowing dynamicthe pressure slope on incoming surface is flow 281 Pa/m. The type The of artificial simulation dynamic pressure slopeflow on incoming surface is sensor 281 Pa/m. sensor lateral is MS5803-07BA, which possess resolution of 0.097 mbar with the digital-analog type ofline artificial lateral line is MS5803-07BA, which possess resolution of 0.097 mbarconversion with the sampling frequency at 512 Hz. The minimum distance written as:distance X digital-analog conversion sampling frequency at 512Xmin Hz.can Thebeminimum can be min

written as:

Sp 9.7 Xmin = = = 0.035 m (14) 9.7 kp 281 = = = 0.035 m (14) 281 Considering that the artificial lateral line system produces the process and circuit arrangement, Considering that the artificial lineeach system produces the process and circuitinarrangement, the sensors are spaced with 50 mmlateral between other. Five columns are arranged the cylinder the sensors are as spaced with surface, shown Figure 6. 50 mm between each other. Five columns are arranged in the cylinder surface, shown as Figure 6.

Sensors 2018, 18, 838

7 of 24

Sensors 2018, 18, x FOR PEER REVIEW

7 of 24

(a)

(b)

Figure 6. Artificial lateral line system. (a) Sensor distribution; (b) 3D modeling. This model includes Figure 6. Artificial lateral line system. (a) Sensor distribution; (b) 3D modeling. This model includes the shell, sensors and embedded hardware. the shell, sensors and embedded hardware.

In this paper, we obtain the pressure distribution cloud of the carrier by hydrodynamic In this paper, obtainand the find pressure cloud oflocation the carrier by hydrodynamic simulation simulation of thewe carrier the distribution pressure sensitive of the carrier from the pressure ofcloud the carrier and find the pressure sensitive location of the carrier from the pressure cloud graph, graph, such as the pressure maximum point, the pressure minimum point and the such static aspressure the pressure maximum point, the pressure minimum point and the static pressure zero point. These zero point. These locations are the best locations for sensor placement. No matter how the locations aresize the best locations sensor itplacement. No matter howway the shape and of location the carrier shape and of the carrier for changes, can be analyzed in this to find thesize best for changes, it can be analyzed in this way to find the best location for sensor placement. sensor placement. 4. Obstacle Sensing Algorithm Based on Simulation 4. Obstacle Sensing Algorithm Based on Simulation Turbulence fluid environment is complex in irregular vortex or gradient flow. In order to obtain Turbulence fluid environment is complex in irregular vortex or gradient flow. In order to obtain an obstacle identification method, a periodic fluid environment formed by Karman vortex streets an obstacle identification method, a periodic fluid environment formed by Karman vortex streets behind an obstacle is utilized as a simulated scene to represent turbulent flow. This environment behind an obstacle is utilized as a simulated scene to represent turbulent flow. This environment can can be reproduced by the experimental sink, which can easily verify the simulation results through be reproduced by the experimental sink, which can easily verify the simulation results through the the experimental data. It can be assumed that the all faced with an obstruction in front of it under experimental data. It can be assumed that the all faced with an obstruction in front of it under certain certain flow velocity. Its surface pressure distribution will differ from the shape, dimension and motion flow velocity. Its surface pressure distribution will differ from the shape, dimension and motion of of the obstacles, which can be features to obtain sensing algorithm. the obstacles, which can be features to obtain sensing algorithm. 4.1. Simulation of Static Obstacle 4.1. Simulation of Static Obstacle The simulation settings are shown in Figure 7. Only one obstacle was located in the flow The simulation settings are shown Figurethrough 7. Onlyaone obstacle was located in kind the flow field field during each simulation. When flowin passed certain object under some of fluid during each simulation. When flow passed through a certain object under some kind of fluid environment, a column of regular vortex will be formed behind the object called Karman vortex streets. environment, a column of regular vortexformula will be isformed behind the object called Karman vortex The vortex shedding frequency calculation as follows: streets. The vortex shedding frequency calculation formula is as follows:   fd 19.7 fd = 0.198 1 − 19.7  (15) St = St V   0.198 1  Re  (15)

V



Re 

where St is for the Strauss number, it is maintained at 0.21 when Reynolds number is in the range of where St is for the Strauss number, it is maintained at 0.21 when Reynolds number is in the range of 300 to 105 .5 A regular Karman vortex streets occurred when Reynolds number in the range of 60–105 .5 300 to 10 . A regular Karman vortex streets occurred when Reynolds number in the range of 60–10 . The simulated surface pressure results are shown in Figures 8–11. The X-axis is a grid node sequence The simulated surface pressure results are shown in Figures 8–11. The X-axis is a grid node sequence distributed along the surface of the carrier, with a total of 182 nodes, the 91st grid node is the forefront distributed along the surface of the carrier, with a total of 182 nodes, the 91st grid node is the of the head; the Y-axis is a time series. According to the simulation results, 10–20 s, 20–40 s, 30–60 s forefront of the head; the Y-axis is a time series. According to the simulation results, 10–20 s, 20–40 s, range; Z-axis pressure value. 30–60 s range; Z-axis pressure value.

Sensors 2018, 18, 838

8 of 24

Sensors 2018, 18, x FOR PEER REVIEW

8 of 24

D=50mm D=100mm D=200mm

D=100mm D=200mm

V=0.1m/s

V1=0.1m/s

750mm

(a)

(b)

Figure 7. 7. Simulation Simulation of of static static obstacle. obstacle. (a) (a) Cylindrical Cylindrical obstructions; obstructions; (b) (b) Square Square obstructions. obstructions. The The circle circle Figure represents aa circular circular obstacle, obstacle, the the square square represents represents aa square square obstacle, obstacle, and and the the oval oval shape shape represents represents represents a carrier. a carrier.

The figures showed that at a certain grid node, all pressure curves were superposition of a The figures showed that at a certain grid node, all pressure curves were superposition of a direct direct current component with a periodic component, which conformed to a second order Fourier current component with a periodic component, which conformed to a second order Fourier polynomial. polynomial. In the periodic component, the period of dynamic and static pressure under same In the periodic component, the period of dynamic and static pressure under same obstacle was similar. obstacle was similar. By means of de-mean fast Fourier transform, the frequency domain was By means of de-mean fast Fourier transform, the frequency domain was obtained. There were three obtained. There were three main frequency peaks. In the same simulated environment, the main frequency peaks. In the same simulated environment, the amplitude of once frequency at amplitude of once frequency at stagnation point was the highest, the second and the third frequency stagnation point was the highest, the second and the third frequency amplitude gradually decreased. amplitude gradually decreased. While the dimension of the obstacle increased, all the frequency While the dimension of the obstacle increased, all the frequency amplitude increased. Taking once amplitude increased. Taking once main frequency as a simulated characteristic frequency, the main frequency as a simulated characteristic frequency, the comparison for the theoretical frequency comparison for the theoretical frequency of Karman Vortex shedding to the simulated frequency is of Karman Vortex shedding to the simulated frequency is shown in the Table 3. The feature size shown in the Table 3. The feature size of a square obstacle is the diameter of its circumcircle. When of a square obstacle is the diameter of its circumcircle. When the length of square obstacle was the length of square obstacle was 200 mm, the carrier entered in the obstacle vortex separation zone 200 mm, the carrier entered in the obstacle vortex separation zone and prevented the vortex shedding. and prevented the vortex shedding. So, a stable flow field was formed with no characteristic So, a stable flow field was formed with no characteristic frequency in pressure. frequency in pressure. Table 3. Comparison of the theoretical frequency of Karman Vortex shedding with the simulated Table 3. Comparison of the theoretical frequency of Karman Vortex shedding with the simulated frequency. frequency. Obstacle Dimensions Round Square size/mm 50 Round 100 200 300 100Square (141) Obstacle Feature Dimensions Theoretical frequency/Hz 0.42 0.21 0.105 0.07 0.141 Feature size/mm 50 100 200 300 100 (141) Simulation frequency/Hz 0.42 0.4918 0.165 0.07 0.096 0.149 Theoretical frequency/Hz 0.21 0.298 0.105 0.141 Simulation frequency/Hz

0.4918

0.298

0.165

0.096

0.149

Table 3 showed that an obstacle in the uniform flow field caused an obvious frequency characteristics in surface pressure distribution and the first main frequency of an artificial lateral line Table 3 showed that an obstacle in the uniform flow field caused an obvious frequency system was closed to the Karman Vortex shedding frequency, which can help estimate the obstacle characteristics in surface pressure distribution and the first main frequency of an artificial lateral feature. The surface pressure waveform could distinguish shape of obstacle as Figures 12 and 13 line system was closed to the Karman Vortex shedding frequency, which can help estimate the obstacle shown. feature. The surface pressure waveform could distinguish shape of obstacle as Figures 12 and 13 shown. In the dynamic pressure of circular obstacle, the peak pressure interval is relatively short and In the dynamic pressure of circular obstacle, the peak pressure interval is relatively short and the peak pressure is relatively high but the peak interval time of the square obstacle is relatively the peak pressure is relatively high but the peak interval time of the square obstacle is relatively long. long. It can be seen from the analysis of static pressure cloud chart that the pressure value of square It can be seen from the analysis of static pressure cloud chart that the pressure value of square obstacle obstacle in the straight transitional zone is larger and there are more red areas in the cloud image. in the straight transitional zone is larger and there are more red areas in the cloud image. However, However, in the pressure cloud chart of the circular obstacle, there are more green areas and the in the pressure cloud chart of the circular obstacle, there are more green areas and the pressure values pressure values is smaller. is smaller.

Sensors 2018, 18, x FOR PEER REVIEW Sensors 2018, 18,18, 838 FOR PEER REVIEW Sensors 2018, Sensors 2018, 18, xx FOR PEER REVIEW

(a) (a) (a)

9 of 24 9 of 2424 of 99 of 24

(b) (b) (b)

Figure 8. Cylindrical obstructions with diameter of 50 mm. (a) Dynamic pressure; (b) Static pressure. Figure 8. 8. Cylindrical Cylindrical obstructions obstructions with with diameter diameter of 50 mm. mm. (a) (a) Dynamic Dynamic pressure; pressure; (b) (b) Static Static pressure. pressure. Figure Figure 8. Cylindrical obstructions with diameter ofof5050mm. (a) Dynamic pressure; (b) Static pressure.

(a) (a) (a)

(b) (b) (b)

(a) (a) (a)

(b) (b) (b)

Figure 9. Cylindrical obstructions with diameter of 100 mm. (a) Dynamic pressure; (b) Static Figure 9. 9. Cylindrical Cylindrical obstructions obstructions with with diameter diameter of of 100 100 mm. mm. (a) (a) Dynamic Dynamic pressure; pressure; (b) (b) Static Static Figure pressure. Figure 9. Cylindrical obstructions with diameter of 100 mm. (a) Dynamic pressure; (b) Static pressure. pressure. pressure.

Figure 10. Cylindrical obstructions with diameter of 200 mm. (a) Dynamic Pressure; (b) Static Pressure. Figure 10.10.Cylindrical 200 mm. (b) Figure Cylindricalobstructions obstructionswith withdiameter diameterofof of 200 mm.(a) (a)Dynamic DynamicPressure; Pressure; (b)Static StaticPressure. Pressure. Figure 10. Cylindrical obstructions with diameter 200 mm. (a) Dynamic Pressure; (b) Static Pressure.

Sensors 2018, 18, x FOR PEER REVIEW

10 of 24

Sensors 2018, 18, 838

10 of 24

Sensors 2018, 18, x FOR PEER REVIEW Sensors 2018, 18, x FOR PEER REVIEW

10 of 24 10 of 24

(a)

(b)

(a) (b) (a) (b) Figure 11. Square obstructions with side length of 100 mm. (a) Dynamic Pressure; (b) Static Pressure. Figure 11. Square obstructions with side length of 100 mm. (a) Dynamic Pressure; (b) Static Pressure. Figure 11. Square obstructions with side length of 100 mm. (a) Dynamic Pressure; (b) Static Pressure. Figure 11. Square obstructions with side length of 100 mm. (a) Dynamic Pressure; (b) Static Pressure. Dynamic Pressure/Pa 180 160

Grid Sequence Grid Sequence

Grid Sequence

140 120 100

80 60 40 20

Dynamic Pressure/Pa Dynamic Pressure/Pa

180 180 160 160 140 140 120 120 100 100 80 80 60 60 40 40 20 20

12 12 12 10 10 10 88 8 6 6 6 4 44 2 22

45

45 45

50

55 55

50 Time(s) 50 Time(s)

55

60 60

60

Time(s)(a)

(b) (b)

(a)

(a) (b) Figure 12. Square obstructions with side length of 100 mm. (a) Dynamic Pressure; (b) Static Pressure.

Figure 12. Square obstructions with side length of 100 mm. (a) Dynamic Pressure; (b) Static Pressure.

Figure Dynamic Pressure; Pressure;(b) (b)Static StaticPressure. Pressure. Figure12. 12.Square Squareobstructions obstructionswith withside side length length of of 100 mm. (a) Dynamic

(a) (a)

(b) (b)

Figure 13. Cylindrical obstructions with diameter of 100 mm. (a) Dynamic Pressure; (b) Static Pressure. Figure 13. Cylindrical obstructions with diameter of 100 mm. (a) Dynamic Pressure; (b) Static Pressure.

(a)

(b)

4.2. Simulation of Moving Carrier 4.2. Simulation of Moving Carrier Figure 13. Cylindrical obstructions with diameter of 100 mm. (a) Dynamic Pressure; (b) Static Pressure. In 13. thisCylindrical simulation,obstructions the carrier gradually moves to an obstacle along the flow direction Figure with diameter of 100 mm. (a) Dynamic Pressure; (b) Static through Pressure. In this simulation, thein carrier gradually moves to an obstacle along directionisthrough dynamic mesh technology FLUENT. The basic simulated parameter ofthe the flow environment shown dynamic mesh technology in FLUENT. The basic simulated parameter of the environment is shown 4.2. Simulation of Moving Carrier in Figure 14. 4.2. Simulation in Figure 14.of Moving Carrier

In this simulation, the carrier gradually moves to an obstacle along the flow direction through In this simulation, the carrier gradually moves to an obstacle along the flow direction through dynamic mesh technology in FLUENT. The basic simulated parameter of the environment is shown dynamic mesh technology in FLUENT. The basic simulated parameter of the environment is shown in in Figure 14. Figure 14.

Sensors 2018, 18, 838

11 of 24

Sensors 2018, 18, x FOR PEER REVIEW

11 of 24

V=0.1m/s V=0.1m/s

D=100mm D=300mm Figure 14. A schematic of the simulation environment. Figure 14. A schematic of the simulation environment.

An UDF was programmed to define the motion characteristics. Smoothing and re-meshing An UDF programmed to dynamic define the motion characteristics. Smoothing re-meshing methods were was combined to achieve mesh of the moving carrier. The springand constant factor methods were combined to achieve dynamic mesh of the moving carrier. The spring constant was set to 0.6, sizing function was open and re-meshing methods were applied to Local Cell, factor Local was set 0.6, sizing was pressure open anddata re-meshing methods were applied to Local Local Face andtoRegion Face.function The surface simulations form is shown in Figures 15 Cell, and 16. Face When and Region Face. The surface pressure data simulations form is shown in Figures 15 and 16. the carrier moved, the overall trend of the pressure distribution was unchanged. There the carrier moved, the overall trend of the pressure was unchanged. There wereWhen still period features on surface caused by obstacles. Table 4distribution shows frequency domain features were still period features on surface caused by obstacles. Table 4 shows frequency domain features through mean-value Fourier transform. There were still three main frequency peaks but the through mean-value Fourier transform. were still three main frequency the amplitude amplitude of first frequency was not There the highest. The highest amplitudepeaks was but underline which of first frequency was not thefrequency. highest. The highest amplitude was underline which represented represented the characteristic the characteristic frequency. Table 4. The frequency domain features. Table 4. The frequency domain features.

Diameter/mm Diameter/mm 100 300 100 300

Main Frequency Peak/Hz Main Frequency 0.201 Peak/Hz 0.452 0.256 0.513 0.201 0.452

0.05 0.05 0.05 0.05

0.256

0.513

Compared to the static state, the characteristic frequency of the motion state was increasing.

f1 , increasingfrequency f2 the part was of , moving frequency f′. Carrier Assume that thetostatic frequency Compared the static state,was the characteristic motion state wasisincreasing. 0 Assumeatthat the of static was f 1of, increasing partThere was fare: moved speed 0.1 frequency m/s, the length it was 0.4 m. 2 , moving frequency is f . Carrier moved at speed of 0.1 m/s, the length of it was 0.4 m. There are: + 1 1 0.1 1 + 2 s2 s1 (16) ′ = = v + v T1 + T2 = 1 + 1 = 1 + 2 = 1 + 0.1 = 1 + 0.25 2 1 0 f = = = 1 + 2 = f 1 + f 2 = f 1 + 0.4 = f 1 + 0.25 (16) s s T1 T2 0.4 So, the frequency increase under 0.1 m/s was 0.25 Hz. According to Equation (16), the moving So, theinfrequency under 0.1simulation m/s was 0.25 Hz. According Equation (16), the moving frequency theoreticalincrease calculation and can be calculated as to Table 5. frequency in theoretical calculation and simulation can be calculated as Table 5. Table 5. The comparison of theoretical and simulation results when the carrier moving. Table 5. The comparison of theoretical and simulation results when the carrier moving. Theoretical Shedding Frequency/Hz Static Shedding Frequency/Hz Moving Theoretical Diameter/mm 100 0.21 0.46 Static Moving 300 0.07 0.32

Diameter/mm

100 300

0.21 0.07

0.46 0.32

Simulation Frequency/Hz Moving Simulation Frequency/Hz Static Frequency/Hz Moving Moving Simulation Simulation Frequency/Hz 0.298 0.548 0.452 Static Moving 0.096 0.346 0.256 0.298 0.096

0.548 0.346

0.452 0.256

Table 5 indicated that the moving frequency was close to the theoretical frequency. The simulation that the artificial lateral wouldfrequency. change the frequency Table 5 shows indicated thatmovement the moving of frequency was close to theline theoretical The simulation distribution of surface of pressure. The lateral amountline of would changechange was depended on the velocity and shape. shows that movement the artificial the frequency distribution of surface Therefore, when identifying an obstacle, the movement of artificial lateral line itself should be pressure. The amount of change was depended on the velocity and shape. Therefore, when identifying considered. an obstacle, the movement of artificial lateral line itself should be considered.

Sensors Sensors2018, 2018,18, 18,838 FORPEER PEERREVIEW REVIEW Sensors 2018, 18, x xFOR

(a) (a)

1212of ofof24 2424 12

(b) (b)

Figure15. 15.Surface Surfacepressure pressuredistribution distributionwith withmoving movingcarrier carrierwith withdd= =100 100mm. mm.(a) (a)Dynamic DynamicPressure; Pressure; Figure Figure 15. Surface pressure distribution with moving carrier with d = 100 mm. (a) Dynamic Pressure; (b)Static StaticPressure. Pressure. (b) (b) Static Pressure.

(a) (a)

(b) (b)

Figure16. 16.Surface Surfacepressure pressuredistribution distributionwith withmoving movingcarrier carrierwith withdd= =300 300mm. mm.(a) (a)Dynamic DynamicPressure; Pressure; Figure Figure 16. Surface pressure distribution with moving carrier with d = 300 mm. (a) Dynamic Pressure; (b)Static StaticPressure. Pressure. (b) (b) Static Pressure.

4.3.Simulation SimulationofofVibrating VibratingObstacle Obstacle 4.3. 4.3. Simulation of Vibrating Obstacle Thesimulation simulationenvironment environmentwas wasshown shownininthe theFigure Figure17. 17.The Thecarrier carrierwas wasstatic staticwhile whilethe the The The simulation environment was shown in the Figure 17. The carrier was static while the obstacle obstaclevibrated. vibrated.InInthe thestatic staticstate, state,the thetheoretical theoreticalvortex vortexshedding sheddingfrequency frequencyisis0.21. 0.21.The TheFLUENT FLUENT obstacle vibrated. In the static state, the theoretical vortex shedding frequency is 0.21. The FLUENT UDF(User (UserDefined DefinedFunction) Function)function functionisisused usedtotocontrol controlthe theobstacle obstaclevibration. vibration.The Theamplitude amplitudeofof UDF UDF (User Defined Function) function is used to control the obstacle vibration. The amplitude thevibration vibrationwas was0.01 0.01mmand andthe thefrequency frequencywas was0.2 0.2Hz. Hz.The Thevibration vibrationequation equationwas wasasasfollows: follows: the of the vibration was 0.01 m and the frequency was 0.2 Hz. The vibration equation was as follows: (17) 0.01 sin sin22 0.2 (17) ==0.01 0.2 y = 0.01 × sin 2π × 0.2t (17) Theresult resultofofsimulation simulationisisshown shownininFigure Figure18. 18. The The figures indicated that the Karman vortex phenomenonstill stillexisted existedwhen whenthe theobstacle obstaclewas was The figures indicated that the Karman vortex The result of simulation is shown in Figure 18. phenomenon slightly vibrating butthe thethat backthe pressure extremes disappeared.still Because vibration direction slightly but back pressure extremes disappeared. Because vibration direction isis The vibrating figures indicated Karman vortex phenomenon existed when the obstacle perpendicular to the flow, the shedding vortexes possessed a velocity component in the direction perpendicular to the flow, the shedding vortexes possessed a velocity component in the direction of was slightly vibrating but the back pressure extremes disappeared. Because vibration directionof vibration whichweakened thevortex vortexforce force whenthey theymoved moved theend end thecarrier. carrier. Fromthe the vibration which when totothe ofofthe is perpendicular toweakened the flow,the the shedding vortexes possessed a velocity component in the From direction frequency domain, thecharacteristic characteristic frequency was0.199 0.199 Hzwhich which wasless lessthan than staticFrom state. frequency domain, frequency was ititcarrier. atatstatic state. of vibration which the weakened the vortex force when theyHz moved towas the end of the One of the reasons may be that the vibration reduced the frequency of vortex shedding. One of the reasons may be that the vibration reduced the frequency of vortex shedding. the frequency domain, the characteristic frequency was 0.199 Hz which was less than it at static state. One of the reasons may be that the vibration reduced the frequency of vortex shedding.

Sensors 2018, 18, 838

13 of 24

Sensors 2018, 18, x FOR PEER REVIEW Sensors 2018, 18, x FOR PEER REVIEW

13 of 24 13 of 24

V1=0.1m/s V1=0.1m/s

D=100mm D=100mm

750mm 750mm

Figure 17. The schematic of simulation environment. In the picture, a circle represents a circular Figure 17. The schematic of simulation environment. In the picture, a circle represents a circular Figure 17. The schematic of simulation obstacle, carrier.environment. In the picture, a circle represents a circular obstacle, and and the the oval oval represents represents aa carrier. obstacle, and the oval represents a carrier.

(a) (a)

(b) (b)

Figure 18. Surface pressure distribution with static carrier. (a) Dynamic Pressure; (b) Static Pressure. Figure 18. Surface pressure distribution with static carrier. (a) Dynamic Pressure; (b) Static Pressure. Figure 18. Surface pressure distribution with static carrier. (a) Dynamic Pressure; (b) Static Pressure.

When the amplitude of vibrating increased, the surface pressure waveform became distortion, Whenconformed the amplitude of vibrating increased, the surface pressurebut waveform distortion, no longer to a standard second order Fourier polynomial the mainbecame frequency of the Whenconformed the amplitude of vibrating increased, the surface pressure but waveform became distortion, no longer to a standard second order Fourier polynomial the main frequency of the pressure was still 0.199 Hz. High-order components appeared in the frequency domain when no longer conformed to a standard second order Fourier polynomial but the main frequency of pressure frequency was still 0.199 Hz. which High-order components in the frequency domain when vibrating increased, are shown in Table appeared 6. the pressure was still 0.199 Hz. High-order components appeared in the frequency domain when vibrating frequency increased, which are shown in Table 6. vibrating frequency increased, which are shown in Table 6. Table 6. The change of frequency when vibrating frequency increased. Table 6. The change of frequency when vibrating frequency increased. Table 6. The change of frequency when vibrating frequency increased. Vibrating Frequency\Hz Pressure Main Frequency\Hz

Vibrating 0.2 Frequency\Hz Vibrating Frequency/Hz 0.2 0.4

0.4 0.20.6 0.40.6 0.6

Main Frequency\Hz 0.049Pressure 0.199 0.298 0.398 Pressure Main 0.049 0.199 0.298 0.398 0.049 0.149 Frequency/Hz 0.248 0.348 0.049 0.149 0.248 0.348 0.049 0.298 0.398 0.149 0.199 0.248 0.348 0.447 0.0490.149 0.149 0.248 0.348 0.248 0.348 0.447 0.149

0.248

0.348

0.447

From the qualitative analysis, the vibration state of the obstacle can be identified by the From theofqualitative analysis, the vibration state of the obstacle can be identified by the extreme value the back pressure. extreme value of the back pressure. From the qualitative analysis, the vibration stateflow of the obstacle can be identified by the leads extreme In the simulation analysis of two-dimensional field, the movement of the carrier to In the simulation analysis of two-dimensional flow field, the movement of the carrier leads to value of the back the increase of thepressure. frequency characteristics of the pressure signal on the surface, which is positively the increase ofvelocity. the frequency characteristics of the pressure signal on the is positively In to thethe simulation analysis of two-dimensional flowstill field, the ofwhich the carrier leads related The Karman vortex phenomenon exists inmovement the surface, small amplitude vibration related to the velocity. The Karman vortex phenomenon still exists in the small amplitude vibration to the increase of the frequency characteristics of the pressure signal on the surface, which is of the obstacle but the extreme value at the tail of the dynamic pressure disappears and the value of of the obstacle but the extreme value at the tail of the dynamic pressure disappears and the value of positivelypressure related toisthe velocity. Thethan Karman phenomenon still exists the smallreduces amplitude dynamic much smaller that vortex of the stationary obstacle. The in vibration the dynamic of pressure is much thanleads that obstacle. The vibration reducesand the vibration obstacle butsmaller the which extreme valueof thestationary tail of theof dynamic pressure disappears frequency ofthe vortex shedding, toatthe the decrease the characteristic frequency of frequency of vortex shedding, which leads to the decrease of the characteristic frequency of the valueon ofthe dynamic is much than that of the stationary obstacle.ofThe vibration pressure surfacepressure of the carrier. Thesmaller vibration amplitude affects the waveform pressure and pressure on the surface of the carrier. The vibration amplitude affects the waveform of pressure and reduces the frequency of vortex shedding, which leads to the of thefrequency characteristic frequency the vibration frequency leads to high-order components indecrease the pressure domain. The the vibration frequency leads to high-order components in the pressure frequency domain. of pressure on the surface of the carrier. The vibration amplitude affects the waveform of square obstacle is similar to a circle; however, it is necessary to use the diameter of The the square obstacle is similar to a circle; however, it is necessary to use the diameter of the pressure and the vibration frequency leads tofrequency high-order in shedding. the pressure circumscribed circle to calculate the shedding of components Karman vortex By frequency obtaining circumscribed to calculate shedding on frequency of Karmanresults, vortexitshedding. By obtaining the influence ofcircle the change of thethe parameters the experimental can provide the basis the influence of the change of the parameters on the experimental results, it can provide basis for the establishment of the algorithm of identifying the shape, the size and the movementthe state of for the establishment of the algorithm of identifying the shape, the size and the movement state of

Sensors 2018, 18, 838

14 of 24

domain. The square obstacle is similar to a circle; however, it is necessary to use the diameter of the circumscribed circle to calculate the shedding frequency of Karman vortex shedding. By obtaining the influence of the change of the parameters on the experimental results, it can provide the basis for the establishment of the algorithm of identifying the shape, the size and the movement of Sensors 2018, 18, x FOR PEER REVIEW 14 ofstate 24 the obstacle. In the experiment, we set the obstacle and carrier distance as 720 mm and carrier shape as the obstacle. In the experiment, we set the obstacle carrier distance as 720 mmof and carrier shape torpedo type. following experiment, we will and further study the influence these two14 variables Sensors 2018, In 18, the x FOR PEER REVIEW of 24 as torpedo type. In the following experiment, we will further study the influence of these two on the perception of flow field. variables on the perception of flow thegeneral obstacle. In the experiment, we field. set the obstacle and carrier distance as 720 mm and shape A process of environmental perception method by artificial lateral linecarrier was conducted A general process of environmental perception by artificial lateral line was as torpedo type. In the following experiment, we method will further study the influence of conducted these two through above analysis, which is shown as Figure 19. through above analysis, which is shown as Figure 19. variables on the perception of flow field. A general process of environmental perception method by artificial lateral line was conducted through aboveSTART analysis, which is shown as Figure 19. Pressure data Fast Fourier STARTtransform

YES

Pressure data Fast Fourier transform Feature frequency? NO Feature frequency? No obstacle

YES

obstacle

Static obstacle

obstacle Tail extremes？

Static obstacle Triangle waveform recognition YES NO Triangle waveform recognition Square obstacle Circle YES NO obstacle YES

Tail NO extremes？ vibrating Obstacle NO

NO No obstacle END

END

Carrier motion?

YES

vibrating ENDObstacle

SquareEND obstacle

YES

Carrier motion?NO

d=0.21V/(f-V/0.4) YES

d=0.21V/f NO

d=0.21V/(f-V/0.4)

d=0.21V/f

Circle obstacle

END

Figure 19. The process of environmental perception method.

END Figure 19. The END process of environmental perception method.

END

5. Experiments of Artificial Lateral Line Figure 19. The process of environmental perception method.

5. Experiments of Artificial Lateral Line

There were totally 25 pressure sensors on artificial lateral line surface, which distributed as 5. Experiments of Artificial Lateral Line on artificial figure. Microcontrollers, wireless communication modules inside as thefigure. There were totally 25 pressure sensors lateraland linebatteries surface, were whichplaced distributed carrier while charging and powering were outside through two watertight connectors. The whole Microcontrollers, wireless modules and batteries were placedwhich insidedistributed the carrieraswhile There were totallycommunication 25 pressure sensors on artificial lateral line surface, system was sealed bywere O-ring. figure. Microcontrollers, wireless communication modules connectors. and batteriesThe were placed inside the charging and powering outside through two watertight whole system was sealed carrier while charging and powering were outside through two watertight connectors. The whole by O-ring.

5.1. Experiments Design system was sealed by O-ring. The overall control system was divided into the host and the lower machine, the lower 5.1. Experiments Design 5.1. Experiments Design in underwater carrier, the host was a data received software running on machine was running The overallthe control system was divided into the host and the lower machine, the lower machine computers, overall program in Figure 20.the host and the lower machine, the lower The overall control systemis shown was divided into

was running in underwater carrier, the host was a data received software running on computers, machine was running in underwater carrier, the host was a data received software running on the overall program is shown in Figure 20. in Figure controller STMF103ZET6 computers, the overall program is shown Master 20. IIC

The lower machine

The lower machine 2.4G wireless RF communications

2.4G wireless RF communications

The host software

The host software

Pressure sensor

MS5803-07BA

Master Mastercontroller switch

STMF103ZET6 Watertight switch

Pressure sensor Power supply

12v lithium MS5803-07BA battery Watertight switch 12v to 5v

Master switch Step-down module Communication Power supply module Step-down module Communication Communication module module data storage Communication module

12v lithium UTC4432 battery 12v to 5v

IIC Power for Master controller

Power for Master controller 2.4G wireless RF

UTC4432 UTC4432 ‘.csv’ format

2.4G wireless RF

UTC4432

Figure 20. The overall program about control system. data storage ‘.csv’ format

There were totally 25Figure pressure sensors which were numbered according to the Figure 6. The The overall program about control system. Figure 20.20. overall program controloutput system. supply voltage of sensors is 3.3 V,The which is given by about the voltage of mater controller. The sensor resolution is 0.04 mbr when the oversampling ratio is 4096 meanwhile the There were totally 25 pressure sensors which were numbered accordingHz, to the Figure 6. The analog-to-digital response time is 8.2 ms. to the DA conversion time of The the There totally 25 conversion pressure which were according the Figure 6. The supply supplywere voltage of(DA) sensors is 3.3 sensors V, which is given bynumbered theLimited voltage output ofto mater controller. digital sensor, the signal sampling frequency of artificial line system was 9.524 Hz. voltage of sensors is pressure 3.3 which giventhe by the voltage output mater controller. The sensor sensor resolution is V, 0.04 mbr is when oversampling ratiolateral isof 4096 Hz, meanwhile the The host software programmed in C# language and could receive and store pressure data in real analog-to-digital (DA) conversion response time is 8.2 ms. Limited to the DA conversion time of the resolution is 0.04 mbr when the oversampling ratio is 4096 Hz, meanwhile the analog-to-digital time. physical hardwaresignal connection is shown in Figure 21. digitalThe sensor, the pressure sampling frequency of artificial lateral line system was 9.524 Hz. The host software programmed in C# language and could receive and store pressure data in real time. The physical hardware connection is shown in Figure 21.

Sensors 2018, 18, 838

15 of 24

(DA) conversion response time is 8.2 ms. Limited to the DA conversion time of the digital sensor, the pressure signal sampling frequency of artificial lateral line system was 9.524 Hz. The host software programmed in C# language and could receive and store pressure data in real time. The physical hardware is REVIEW shown in Figure 21. Sensors 2018,connection 18, x FOR PEER 15 of 24

Head

Master controlle r

Sensors

Tail

Watertight switch

Wireless module

Step-down module

Battery

Figure 21. The physical hardware connection of lateral line. Figure 21. The physical hardware connection of lateral line.

5.2. Underwater Experiments 5.2. Underwater Experiments This experiment was carried out in an experimental tank. One end of the tank is equipped with This experiment was carried out in an experimental tank. One end of the tank is equipped with a high-power pump for generating a flow. The specific parameters of the experimental sink are a high-power pump for generating a flow. The specific parameters of the experimental sink are shown shown in Table 7. in Table 7. Table 7. The The specific specific parameters parameters of of the the experimental experimental sink. sink. Table 7. Item Item Pool size Water density Pool size Water density Experimental water temperature Experimental Maximum flow rate water temperature Maximum Maximum ideal flow velocityflow rate Maximum ideal flow velocity

Parameters Parameters 1 W × 1.14 (H)(m) 1.0(H)(m) × 103 kg/m3 1 W × 1.14 3 1.0 × 10 kg/m 18 3°C 18 ◦ C 0.8 m3/s 0.8 m3 /s0.8 m/s 0.8 m/s

Artificial lateral line was placed in the middle of the tank with wireless communication device inside. In order to line ensure integrity wireless the wireless artificial communication system was installed Artificial lateral wasthe placed in the of middle of thesignal, tank with device underwater for 20 cm to 30 The experiment shown in Figuresystem 22. was installed underwater inside. In order to ensure thecm. integrity of wirelesswas signal, the artificial order collect characteristic as a control group, for 20Incm to 30tocm. Thestatic experiment was shown in Figure 22. the artificial carrier was placed in still waterInand recorded pressure data before generating flow. The characteristic related to order to collect static characteristic as a controla group, the static artificial carrier waswas placed in still the current depth,pressure water data temperature, inherent a characteristics the sensors other water and recorded before generating flow. The static of characteristic wasand related to environmental factors. experimental data need to be ofprocessed after the static the current depth, water The temperature, inherent characteristics the sensors and removing other environmental characteristics. The experiments conducted according researchthe goals in characteristics. Section 2 and factors. The experimental data were need to be processed after to removing static simulation parameters. The experiments were conducted according to research goals in Section 2 and simulation parameters.

Sensors 2018, 18, 838

16 of 24

Sensors 2018, 18, x FOR PEER REVIEW

16 of 24

Figure 22. The sink experiment of lateral line. Figure 22. The sink experiment of lateral line.

6. Experimental Analysis of Artificial Lateral Line 6. Experimental Analysis of Artificial Lateral Line Experimental data could verify simulation results and obtain the method of underwater fluid Experimental data could verify simulation results and obtain the method of underwater fluid detection. Due to improper installation, sensors 7, 10, 12, 16 and 18 were unrecoverable mechanical detection. Due to improper installation, sensors 7, 10, 12, 16 and 18 were unrecoverable mechanical damage, there remain 20 available sensors. damage, there remain 20 available sensors. 6.1. Hydrostatic Correction 6.1. Hydrostatic Correction The pressure pressure data data of of the the No. No. 11 sensor sensor was was shown shown in in Figure Figure 23a, 23a, the the frequency frequency domain domain obtained obtained The by fast Fourier transform was shown in Figure 23b. There was an obvious static characteristic in the by fast Fourier transform was shown in Figure 23b. There was an obvious static characteristic original signal,signal, because the highest amplitude appeared at 0at Hz frequency in the original because the highest amplitude appeared 0 Hzand andother other high high frequency components was 0 Pa. Through low-pass filtering to remain only 0 Hz, the pressure data was was components was 0 Pa. Through low-pass filtering to remain only 0 Hz, the pressure data converted to time domain by inverse Fourier transform. Figure 23c was the hydrostatic pressure converted to time domain by inverse Fourier transform. Figure 23c was the hydrostatic pressure after afterprocessing, the processing, was averaged. The hydrostatic of all underwater sensors was the whichwhich was averaged. The hydrostatic pressurepressure of all underwater sensors was obtained obtained by this method. by this method.

Sensors 2018, 18, 838

17 of 24

Sensors 2018, 18, x FOR PEER REVIEW

17 of 24

Pressure Value/Pa

5

1.0259

（a）Original Pressure Signal

x 10

1.0258 1.0257 1.0256

0

5

10

15 20 25

30 35 40

45 50 55

60 65 70

75

Time(s) 7

x 10

Pressure/Pa

10

（b）Pressure Signal Frequency Domain Distribution

5 0

0

0.5

1

1.5

2

2.5

3

3.5

4

Frequency/Hz Pressure Value/Pa

5

1.0258

（c）Filtered Pressure Signal

x 10

1.0257 1.0257

0

5

10

15 20 25

30 35 40

45 50 55

60 65 70

75

Time(s) Figure Figure 23. 23. The The process process of of the the No. No. 11 sensor sensor pressure pressure data. data.

6.2. Velocity Estimation 6.2. Velocity Estimation The experimental data subtracted hydrostatic pressure was compared with the static pressure The experimental data subtracted hydrostatic pressure was compared with the static pressure in the simulation. For uniform flow field, there were three methods, respectively, stagnation pressure in the simulation. For uniform flow field, there were three methods, respectively, stagnation fitting, static pressure fitting, Bernoulli method, for turbulent flow field, there was Karman vortex pressure fitting, static pressure fitting, Bernoulli method, for turbulent flow field, there was Karman method. vortex method. Stagnation pressure fitting method: Stagnation pressure fitting method: When the carrier was facing the flow, the center of the head was stagnation point (point A in When the carrier was facing the flow, the center of the head was stagnation point (point A in simulation, sensor 25 on carrier). According to the simulation results, the relationship between simulation, sensor 25 on carrier). According to the simulation results, the relationship between pressure pressure and flow velocity at point A was as follows: and flow velocity at point A was as follows:

Ps  485.9v2  5.033v  0.41

Ps = 485.9v2 + 5.033v − 0.41 According to the experimental data, the actual fitting curve is: According to the experimental data, the actual fitting curve is: 2

Pt  172.2v 135.75v 17.55

(18) (18)

(19) Pt = 172.2v + 135.75v − 17.55 (19) The fitting degree between the simulation curve and the experimental data was 0.9675, while the actual fit degree was 0.9755. The fitting degree between the simulation curve and the experimental data was 0.9675, Figure 24 showed thatwas the0.9755. simulation results fit well to experimental data under low velocity, while the actual fit degree whileFigure the simulation was higher than the actual data the to case of high velocity. simulated 24 showed that the simulation results fitinwell experimental data Since underthe low velocity, fluid was set to ideal liquid water, the experimental water was turbid. The density and viscosity of while the simulation was higher than the actual data in the case of high velocity. Since the simulated the ideal fluid are all constant. But for turbid water, these are all changing. As the flow rate increases, fluid was set to ideal liquid water, the experimental water was turbid. The density and viscosity of the waterfluid quality becomes more thewater, internal friction water increases and rate the increases, dynamic the ideal are all constant. Butturbid, for turbid these are allinchanging. As the flow viscosity of water increases. As a result, the loss along the water flow increases and the pressure loss the water quality becomes more turbid, the internal friction in water increases and the dynamic getting bigger and bigger. But this situation is not in the simulation, so the simulation was higher viscosity of water increases. As a result, the loss along the water flow increases and the pressure loss than the actual data in the case of high velocity. At the same time, experimental instruments such as the gun, the sensor itself is also present measurement error. According to the cumulative error of 2

Sensors 2018, 18, 838

18 of 24

Pressure Value(Pa)

getting bigger and bigger. But this situation is not in the simulation, so the simulation was higher than Sensors 2018,data 18, xin FOR of 24 the actual thePEER caseREVIEW of high velocity. At the same time, experimental instruments such as the18gun, the sensor itself is also present measurement error. According to the cumulative error of experimental experimental instruments, as the flow rate increases, the error will also increase. This is also why the instruments, as the flow rate increases, the error will also increase. This is also why the simulation simulation data is higher than the experimental data. data is higher than the experimental data.

Figure 24. The results fit of the simulation. Figure 24. The results fit of the simulation.

Static pressure fitting method: Static pressure fitting method: Takesensors sensors 9, 9, 14, 14, 21, 8, 8, 11,11, 19,19, 20 20 to represent the Take 21, 22 22 to to represent representthe thelateral lateralline lineAAand andsensors sensors to represent lateral line B. The corresponding location in simulation was point C (see Figure 25). The theoretical the lateral line B. The corresponding location in simulation was point C (see Figure 25). The theoretical curvewas wasfitting fittingcurve curveof ofpoint pointC, C,which whichcould couldbe beexpressed expressedas: as: curve Ps  165.9v2  6.144v  0.7572 (20) Ps = −165.9v2 + 6.144v − 0.7572 (20) Average value of each lateral line sensor was their mean static pressure. The fitting degree Average value curve of each lateral line sensor their mean static pressure. The fitting between theoretical and experimental datawas were 0.9657 (lateral line A) and 0.9476 (lateraldegree line B). between theoretical were 0.9657 According to sensorcurve data,and the experimental actual fitting data equations were: (lateral line A) and 0.9476 (lateral line B). According to sensor data, the actual fitting equations were: PA  262.4v2  24.63v  1.08 (21) 2 PA P = −262.4v 24.63v 1.08 365.3v2 −  41.02 v − 0.608

(21) (22)

B

2

= −365.3v 41.02v − 0.608 The actual fitting degree werePB0.9673 (lateral+line A) and 0.9486 (lateral line B).

(22)

The actual fitting degree were 0.9673 (lateral line A) and 0.9486 (lateral line B). Bernoulli method: Bernoulli method: According to Bernoulli’s law, the fluid’s stagnation pressure is equal to the sum of static According to Bernoulli’s law, the fluid’s stagnation pressure is equal to the sum of static pressure pressure and dynamic pressure, which can be written as: and dynamic pressure, which can be written as:

where

0

v 2  p0  p fs   V 2 (23) p0  p fs   Vt  u u  2 t 2 p0 − p f s ρV2 → Vt = (23) p0 = p f s + represents stagnation pressure, 2 represents staticρpressure, ρ is fluid density and is

the theoretical velocity. In the pressure, experiment, sensor No.static 25 collected pressureand andV the where p0 represents stagnation p f s represents pressure,stagnation ρ is fluid density t is average value taken by lateral line A and B was static pressure. It was assumed that representing the theoretical velocity. In the experiment, sensor No. 25 collected stagnation pressure and the average actualtaken flowby velocity measured thestatic flowpressure. meter. Because of the system there should be a value lateral line A and Bbywas It was assumed that Ve error, representing actual flow correction factor βby asthe follows: velocity measured flow meter. Because of the system error, there should be a correction factor β as follows: β =

Ve

Vt

Ve Ve  V V=t r 2  p  pe   fs  0 

(24) (24)

2 p0 − p f s /ρ The fitting degree between Ve and Vt was 0.9925. Figure 26 showed that β was the0.9925. actual flow velocity can be expressed as: The fitting degree between Ve 0.9072, and Vt so was





Ve  0.9072Vt  0.02518  0.9072 2 p25  p ABCD /1000  0.02518  0.0405

p

25



 p ABCD  0.02518

(25)

Sensors 2018, 18, 838

19 of 24

Figure 26 showed that β was 0.9072, so the actual flow velocity can be expressed as: p = 0.9072Vt − 0.02518 =p 0.9072 2( p25 − p ABCD )/1000 − 0.02518 Sensors Sensors2018, 2018,18, 18,xxFOR FORPEER PEERREVIEW REVIEW = 0.0405 ( p25 − p ABCD ) − 0.02518 Ve

(25) 19 19of of24 24

where pressure sensor No. 25, isisthe average value taken lateral line and p2525isisthe ppABCD where pp25 isthe the pressure of sensor No. 25,p ABCD the average value taken by lateral line and where pressure ofof sensor No. 25, is the average value taken byby lateral line AAA and B ABCD static pressure. Bwas static pressure. Bwas was static pressure. Latera LateraLine LineAA

20 20

Pressure Value(Pa) Value(Pa) Pressure

Lateral LateralLine LineBB

22

PPSS=-165.9V =-165.9V+6.144v+0.757 +6.144v+0.757

00

Theoretical Theoreticalcarve carve

-20 -20

BBSide SideLine LineFitting Fitting

-40 -40

AASide SideLine LineFitting Fitting

-60 -60

2

2 PPB=-356.3V =-356.3V+41.02v-0.608 +41.02v-0.608 B

-80 -80

-100 -100 -120 -120

2

2 PPA=-262.4V =-262.4V-24.63v-1.08 -24.63v-1.08 A

-140 -140 -160 -160 00

0.1 0.1

0.2 0.2 0.3 0.3 0.4 0.4 0.5 0.5 Flow FlowVelocity(m/s) Velocity(m/s)

0.6 0.6

Figure Figure25. 25.The Thepressure pressurecurve curvefitting. fitting. Figure 25. The pressure curve fitting. 0.6 0.6

Experimental Experimentaldata data Curve Curvefitting fitting

Actual Flow Flow Velocity Velocity Ve(m/s) Ve(m/s) Actual

0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2

VVee=0.9072V =0.9072Vt-0.02518 -0.02518 t

0.1 0.1 0.1 0.1

0.2 0.2

0.3 0.3

0.4 0.4

0.5 0.5

0.6 0.6

Theoretical TheoreticalFlow FlowVelocity VelocityVt(m/s) Vt(m/s)

Figure Figure26. 26.The Thecurve curvefitting fittingof ofthe theexperimental experimentaldata. data. Figure 26. The curve fitting of the experimental data.

Karman Karmanvortex vortexmethod: method: Karman vortex method: For Forturbulent turbulentfield, field,the theobstacle obstaclediameter diameterwas wasfixed fixedat at100 100mm. mm.As Asthe thevelocity velocityincreased, increased,the the For turbulent field, the obstacle diameter was fixed at 100 mm. As the velocity increased, Karman Karman vortex vortex appeared appeared clearly. clearly. Due Due to to the the complexity complexity of of the the fluid fluid environment, environment, the the current current the Karman vortex appeared clearly. Due to the complexity of the fluid environment, the current velocity velocitycannot cannotbe beobtained obtainedby bythe theBernoulli Bernoullimethod. method.Due Dueto tothe theformula formulaof ofKarman Karmanvortex vortexshedding shedding velocity cannot be obtained by the Bernoulli method. Due to the formula of Karman vortex shedding frequency, frequency, in in case case of of known known obstacle obstacle dimension, dimension, the the current current flow flow velocity velocity can can be be calculated calculated frequency, in case of known obstacle dimension, the current flow velocity can be calculated according according accordingto tothe thecharacteristic characteristicfrequency frequencyof ofthe thepressure pressuresignal. signal. to the characteristic frequency of the pressure signal. The Theamplitude-frequency amplitude-frequencycharacteristic characteristicof ofthe thesensor sensorNo. No.22shown shownin inFigure Figure27 27was wasobtained obtainedby by The amplitude-frequency characteristic of the sensor No. 2 shown in Figure 27 was obtained by fast fastFourier Fouriertransform. transform.ItItcan canbe beseen seenfrom fromthe thefigure figurethat thatwith withthe theincrease increaseof ofthe theflow flowvelocity, velocity,the the fast Fourier transform. It can be seen from the figure that with the increase of the flow velocity, average average amplitude amplitude of of the the data data increased increased and and the the higher higher amplitude amplitude of of frequency frequency appeared. appeared. the average amplitude of the data increased and the higher amplitude of frequency appeared. According Accordingto tothe theKarman Karmanvortex vortexformula, formula,the therelationship relationshipbetween betweenflow flowvelocity velocityand andthe thefrequency frequency when whenthe theobstacle obstaclediameter diameterisis100 100mm mmwas: was:

SSt t 

fdfd f f 0.1 0.1  0.21 0.21 VV0.476 0.476f f VV VV

(26) (26)

Sensors 2018, 18, 838

20 of 24

According to the Karman vortex formula, the relationship between flow velocity and the frequency when the obstacle diameter is 100 mm was: St =

fd f × 0.1 = = 0.21 → V = 0.476 f V V

(26)

For the actual pressure signal, there is more extreme pressure among different frequency ranges. Sensors the 2018,high 18, x frequency FOR PEER REVIEW 20 of 24 Taking pressure pmax as a criterion, when p(f ) > pmax , the corresponding frequency was regarded as the characteristic frequency, pmax was written as: For the actual pressure signal, there is more extreme pressure among different frequency ranges. Taking the high frequency pressure max as a criterion, when p(f) > pmax, the corresponding p + 1.5σ (27) pmaxp= frequency was regarded as the characteristic frequency, pmax was written as: The actual fitting relationship was as follows: pmax  p  1.5 (27) V= 0.26 f + 0.067 The actual fitting relationship was as follows:

(28)

V  0.26 f in  0.067 The fitting effect of each method is summarized Table 8.

(28)

The fitting effect of each Table method is summarized 8. 8. The fitting degreein of Table each method. Table 8. The fitting degree of each method. Velocity Estimated Method Fit Degree

Flow Field

Flow Field Uniform Uniform turbulent turbulent

Velocity pressure Estimated Method stagnation fitting static pressure fitting stagnation pressure fitting Bernoulli method static pressure fitting Bernoulli method Karman vortex method Karman vortex method

0.9755 Fit Degree 0.94–0.96 0.9755 0.9925 0.94–0.96

0.9925 0.9893

0.9893

25

Pressure(Pa)

20 v=0.090 v=0.195 v=0.267 v=0.433 v=0.526 v=0.695

15 10 5 0 0.8

0.6

0.4

Flow Velocity(m/s)

0.2

0

0

1

2

3

4

Frequency(Hz)

Figure Figure27. 27.The Theamplitude-frequency amplitude-frequencycharacteristic characteristicofofthe thesensor sensorNo. No.2.2.

6.3. 6.3.Attitude AttitudePerception Perception According AccordingtotoFigure Figure28, 28,ititcan canbebeseen seenthat thatthe theangle anglebetween betweenflow flowdirection directionand andartificial artificialcarrier carrier would incident flow flow surface and countercurrent surface, resulting the pressure distribution wouldproduce produce incident surface and countercurrent surface, inresulting in the pressure difference, which can be foundation estimate the lateral line attitude. Takingline the attitude. differenceTaking of lateral distribution difference, which cantobe foundation to estimate the lateral the line A, B corresponding position (6–8, 13–11, 21–19, 22–20) to13–11, remove21–19, the impact by turbulence, difference of lateral line A, B corresponding position (6–8, 22–20)caused to remove the impact pressure difference linear fitting degree waslinear shownfitting in Table 9. was shown in Table 9. caused by turbulence, pressure difference degree Table 9. Pressure difference linear fit degree. Sensor Pair 6–8 13–11 21–19 22–20

Fitness V = 0.3 m/s 0.9856 0.9282 0.9643 0.8534

V = 0.5 m/s 0.9917 0.9336 0.9811 0.9538

Sensors 2018, 18, 838

21 of 24

Table 9. Pressure difference linear fit degree. Fitness Sensor Pair

V = 0.3 m/s

V = 0.5 m/s

0.9856 0.9282 0.9643 0.8534 0.9328

0.9917 0.9336 0.9811 0.9538 0.965

6–8 13–11 21–19 22–20 Mean Sensors 2018, 18, x FOR PEER REVIEW

21 of 24 400 22-20 21-19 13-11 6-8

400 300

v=0.3m/s Pressure Difference（Pa）

v=0.5m/s Pressure Difference（Pa）

500

200 100 0 -100 -200 -300 -400 -50

0

50

22-20 21-19 13-11 6-8

300

200

100

0

-100

-200

-300 -50

Angle（°）

0

50

Angle（°）

(a)

(b)

Figure 28. The results of the simulation. (a) Pressure Difference with v = 0.5 m/s; (b) Pressure Figure 28. The results of the simulation. (a) Pressure Difference with v = 0.5 m/s; (b) Pressure Difference Difference with v = 0.3with m/s.v = 0.3 m/s.

From the Table 9 we can see that the average fitting degree of 0.9 or more, it indicated that the the Table 9 we can see that the average fitting degree of 0.9 or more, it indicated that angleFrom and the pressure difference were in a linear relationship, which was as follows: the angle and the pressure difference were in a linear relationship, which was as follows:

 P  3.3 A  17.55 V  0.1 P = −3.3A + 17.55(V = 0.1)  P  5.9 A  42.35 V  0.3 5.9A + 42.35(V = 0.3) P = − 6.6 A  15.66 V  0.5    PP= − 6.6A − 15.66(V = 0.5)   

(29) (29)

where P is the pressure difference, A is the angle between flow direction and artificial carrier. where P is the pressure difference, A is the angle between flow direction and artificial carrier. 6.4. Obstacle Identification 6.4. Obstacle Identification When there was a significant peak in the frequency domain, it could be determined that there When there was a significant peak in the frequency domain, it could be determined that there were obstacles in flow field. In case of known flow velocity, the Karman vortex shedding formula were obstacles in flow field. In case of known flow velocity, the Karman vortex shedding formula can can estimate obstacle size with characteristic frequency. According to simulation results, the feature estimate obstacle size with characteristic frequency. According to simulation results, the feature size size of square obstacle was its circumscribed circle diameter. The summary data were shown in of square obstacle was its circumscribed circle diameter. The summary data were shown in Table 10. Table 10. The estimated value was generally smaller than the actual obstacle size, the average error The estimated value was generally smaller than the actual obstacle size, the average error rate was rate was about 22.72%. about 22.72%. The neural network algorithm was used to estimate the shape of the obstacle in a turbulent The neural network algorithm was used to estimate the shape of the obstacle in a turbulent environment. The output of the circular obstacle was set to 0, while the square obstacle was 1. The environment. The output of the circular obstacle was set to 0, while the square obstacle was 1. experimental data was input in trained model to identify the obstacle shape. Take k1 = 15, k2 = 5. The The experimental data was input in trained model to identify the obstacle shape. Take k1 = 15, k2 = 5. transfer function between the three layers uses the logarithmic S-type transfer function, the training The transfer function between the three layers uses the logarithmic S-type transfer function, the training function is based on the train-Berger-Marquardt reverse propagation method, the learning function function is based on the train-Berger-Marquardt reverse propagation method, the learning function uses the BP learning rule, Performance analysis function uses mean square error performance uses the BP learning rule, Performance analysis function uses mean square error performance analysis, analysis, network topology is shown in Figure 29. network topology is shown in Figure 29. Table 10. The summary data of simulation results. Characteristic Frequency (Hz) 0.667 1.361 2.140

Velocity (m/s) 0.482 0.433 0.413

Calculated Size (mm) 151.7 66.8 40.5

Actual Size (mm) D200 D100 D50

Error Rate 24.15% 33.2% 19%

Sensors 2018, 18, 838

22 of 24

Table 10. The summary data of simulation results. Characteristic Frequency (Hz)

Velocity (m/s)

Calculated Size (mm)

Actual Size (mm)

Error Rate

0.667 1.361 2.140 0.477 0.918

0.482 0.433 0.413 0.534 0.492

151.7 66.8 40.5 235.1 112.5

D200 D100 D50 A200 (282.8) A100 (141.4)

24.15% 33.2% 19% 16.86% 20.43%

After the optimized training model, the recognition rate of square obstacle improved to 95.3%, the recognition rate of circular obstacle was 98.9%. The output of network is shown in Figure 30. Sensors 2018, 18, x FOR PEER REVIEW 22 of 24 Sensors 2018, 18, x FOR PEER REVIEW

22 of 24

Figure 29. The network topology. In the picture, green indicates the input and output layers, purple Figure 29. The network topology. In the picture, green indicates the input and output layers, purple indicates the hidden layer, w is the weight, b is the offset, and  indicates the activation function. Figure The network In the picture, green indicates the input and output layers, purple indicates the29. hidden layer, wtopology. is the weight, b is the offset, and ∼ indicates the activation function. indicates the hidden layer, w is the weight, b is the offset, and  indicates the activation function.

(a) (a)

(b) Figure 30. The output of network. (a) Square obstacle(b) identification; (b) Circle obstacle identification. Figure 30. The output of network. (a) Square obstacle identification; (b) Circle obstacle identification.

Figure 30. The output of network. (a) Square obstacle identification; (b) Circle obstacle identification. 7. Conclusions

7.InConclusions this paper, a cylindrical artificial lateral-line carrier combined with biomechanics, bionics, 7. Conclusions fluid mechanics and embedded system is developed. Firstly, by imitating the mechanism of fish In this paper, a cylindrical artificial lateral-line carrier combined with biomechanics, bionics, In canal this paper, a cylindrical artificial carrier combined with biomechanics, bionics, fluid lateral neuromasts, a program of lateral-line artificial lateral-line system embraced with pressure sensors fluid mechanics and embedded system is developed. Firstly, by imitating the mechanism of fish mechanics and embedded system is developed. Firstly, bythe imitating the mechanism of fish lateral is put forward. Through the hydrodynamic simulation, mathematical relationship between lateral canal neuromasts, a program of artificial lateral-line system embraced with pressure sensors canal neuromasts, a program of artificial lateral-line system embraced with pressure sensors is put pressure, flow velocity and shape are obtained and the optimal distribution of the lateral pressure is put forward. Through the hydrodynamic simulation, the mathematical relationship between forward. the the hydrodynamic the mathematical between pressure, flow flow sensor is Through concluded, lateral line simulation, pressure distribution modelsrelationship in a uniform and turbulent pressure, flow velocity and shape are obtained and the optimal distribution of the lateral pressure velocity shape are obtained and the optimal distribution of the lateral pressure sensor is concluded, field areand established. sensor is concluded, the lateral line pressure distribution models in a uniform and turbulent flow the lateral modelsand in athe uniform and turbulent flow are established. Then, line the pressure artificial distribution lateral line system corresponding carrier arefield designed and realized field are established. theoptimal artificialtopology lateral line system andThe the real-time corresponding carrier are and designed and realized basedThen, on the of the sensor. communication data transmission Then, the artificial lateral line system and the corresponding carrier are designed and realized based onthe the carrier optimaland topology of the sensor. The real-time communication and data transmission between the computer processing terminal are realized by using the wireless data based on the optimal topology of the sensor. The real-time communication and data transmission between the module. carrier and the computer processing terminal are realized theflow wireless transmission On this basis, the construction of the artificial lateral by lineusing system field between the carrier and the computer processing terminal are realized by using the wireless data perception of underwater test platform is realized. transmission module. On this basis, the construction of the artificial lateral line system flow field Finally, the validity of the numerical simulation method is verified by the comparison between perception of underwater test platform is realized. the experimental data and the simulation results. The effective methods of flow velocity estimation, Finally, the validity of the numerical simulation method is verified by the comparison between attitude perception and obstacle identification based on the artificial lateral line are formed, which the experimental data and the simulation results. The effective methods of flow velocity estimation, lays a theoretical foundation for the underwater environment perception. attitude perception and obstacle identification based on the artificial lateral line are formed, which In this research, the concept of coupled bionics is introduced into the underwater sensing lays a theoretical foundation for the underwater environment perception.

Sensors 2018, 18, 838

23 of 24

data transmission module. On this basis, the construction of the artificial lateral line system flow field perception of underwater test platform is realized. Finally, the validity of the numerical simulation method is verified by the comparison between the experimental data and the simulation results. The effective methods of flow velocity estimation, attitude perception and obstacle identification based on the artificial lateral line are formed, which lays a theoretical foundation for the underwater environment perception. In this research, the concept of coupled bionics is introduced into the underwater sensing system creatively. It is proposed that the fish lateral line system should be applied to underwater environment perception and navigation, breaking the traditional detected ways based on acoustic, optical and inertial navigation. The method of bionics and motion control is extended to form the knowledge system of underwater vehicle omnidirectional perception and local precise navigation and positioning, which provides the method and technical basis for the application of artificial lateral line system in the field of underwater vehicle. Acknowledgments: This research was supported by the National Natural Science Foundation of China (Grant No. 61540010), Key Science and Technology Program of Shandong Province, China (Grant No. 2014GH Y115032), People’s Livelihood Science and Technology Plan of Qingdao (Grant No. 14-2-3-63-nsh), Shandong Provincial Natural Science Foundation, China (Grant No. ZR201709240210) and UOW VC Fellowship. Author Contributions: G.L. conceived and designed the experiments; M.W., A.W. and S.W. performed the experiments; A.W. and T.Y. analyzed the data; M.W., R.M. and Z.L. contributed materials and analysis tools; G.L., M.W. and Z.L. wrote the paper. Conflicts of Interest: The authors declare no conflicts of interest.

References 1. 2. 3. 4. 5. 6.

7. 8. 9. 10. 11. 12.

13.

Krahe, R.; Maler, L. Neural maps in the electrosensory system of weakly electric fish. Curr. Opin. Neurobiol. 2014, 24, 13–21. [CrossRef] [PubMed] Blaxter, J.H.S.; Fuiman, L.A. The Mechanosensory Lateral Line: Neurobiology and Evolution; Coombs, S., Gorner, P., Munz, H., Eds.; Springer Science & Business Media: Berlin, German, 2012; pp. 481–499. Kanter, M.J.; Coombs, S. Rheotaxis and prey detection in uniform currents by Lake Michigan mottled sculpin. J. Exp. Biol. 2003, 206, 59–70. [CrossRef] [PubMed] Gibbs, M.A. Lateral line receptors: Where do they come from developmentally and where is our research going. Brain Behav. Evol. 2004, 64, 163–181. [CrossRef] [PubMed] Hoekstra, D.; Janssen, J. Non-visual feeding behavior of the mottled sculpin, Cottus bairdi, in Lake Michigan Environ. Biol. Fishes 2006, 12, 111–117. [CrossRef] Yang, Y.; Nguyen, N.; Chen, N.; Lockwood, M.; Tucker, C.; Hu, H.; Bleckmann, H.; Liu, C.; Jones, D.L. Artificial lateral line with biomimetic neuromasts to emulate fish sensing. Bioinspir. Biomim. 2010, 5, 16001. [CrossRef] [PubMed] Fan, Z.; Chen, J.; Zou, J.; Bullen, D.; Liu, C.; Delcomyn, F. Design and fabrication of artificial lateral line flow sensors. J. Micromech. Microeng. 2002, 12, 655–661. [CrossRef] Kroese, A.B.A.; Schellart, N.A.M. Velocity- and acceleration-sensitive units in the trunk lateral line of the trout. J. Neurophys. 1992, 68, 2212–2221. [CrossRef] [PubMed] Pohlmann, K.; Atema, J.; Breithaupt, T. The importance of the lateral line in nocturnal predation of piscivorous catfish. J. Exper. Biol. 2004, 207, 2971–2978. [CrossRef] [PubMed] Bleckmann, H. 3-D ofientation with the octavolateralis system. J. Physiol. Paris 2004, 98, 53–65. [CrossRef] [PubMed] Pandya, S.; Yang, Y.; Liu, C. Biomimetic Imaging of Flow Phenomena. Int. Conf. Acoust. Speech Signal Process. 2007, 2, II-933–II-936. DeVries, L.; Lagor, F.D.; Lei, H.; Tan, X.; Paley, D.A. Distributed flow estimation and closed-loop control of an underwater vehicle with a multi-modal artificial lateral line. Bioinspir. Biomim. 2015, 10, 025002. [CrossRef] [PubMed] Netten, S.M.V. Hydrodynamics of the excitation of the cupula in the fish canal lateral line. Vet. Dermatol. 1991, 18, 324–331. [CrossRef]

Sensors 2018, 18, 838

14. 15. 16.

17. 18. 19. 20.

24 of 24

Chambers, L.D.; Akanyeti, O.; Venturelli, R. A fish perspective: Detecting flow features while moving using an artificial lateral line in steady and unsteady flow. J. R. Soc. Interface 2014, 11. [CrossRef] [PubMed] Netten, S.M.V. Hydrodynamics detection by cupulae in a lateral-line canal: Functional relations between physics and physiology. Biol. Cybern. 2006, 94, 67–85. [CrossRef] [PubMed] Venturelli, R.; Akanyeti, O.; Visentin, F.; Ježov, J.; Chambers, L.D.; Toming, G.; Brow, J.; Kruusmaa, M.; Megill, W.M.; Fiorini, P. Hydrodynamic pressure sensing with an artificial lateral line in steady and unsteady flows. Bioinspir. Biomim. 2012, 7, 36004–36015. [CrossRef] [PubMed] Tan, S. Underwater artificial lateral line flow sensors. Microsyst. Technol. 2014, 20, 2123–2136. Wannenburg, J.; Malekian, R. Physical Activity Recognition from Smartphone Accelerometer Data for User Context Awareness Sensing. IEEE Trans. Syst. Man Cybern. Syst. 2017, 47, 3142–3149. [CrossRef] Tian, Z.; Liu, F.; Li, Z.; Malekian, R.; Xie, Y. The Development of Key Technologies in Applications of Vessels Connected to the Internet. Symmetry 2017, 9, 211. [CrossRef] Yu, J.; Malekian, R.; Chang, J.; Su, B. Modeling of whole-space transient electromagnetic responses based on FDTD and its application in the mining industry. IEEE Trans. Ind. Inform. 2017, 13, 2974–2982. [CrossRef] © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).