34 th International Syposiu on Autoation and Rootics in Construction (ISARC 2017) Developent of UAV Indoor Flight Technology for Building Equipent Works H. Toita a, T. Takaatake a, S. Sakaoto a, H. Arisui, S. Kato and Y. Ohgusu a Shinryo Corporation, Japan National Institute of Advanced Industrial Science and Technology, Japan E-ail: toita.hr@shinryo.co, takaatake.ta@shinryo.co, sakaoto.sh@shinryo.co, h-arisui@aist.go.jp, shin.kato@aist.go.jp, yuji.ogusu@aist.go.jp Astract In the Japanese construction industry, An UAV has egun to e used for outdoor works. On the other hand, ost of uilding equipent works is indoor work, often working in touch with the surrounding environent near the ceiling. Air volue easureent is an exaple of uilding equipent works near the ceiling. Conventionally, a worker gets on a teporary scaffold to approach the ceiling at the high position and easures the air volue just under the diffuser installed on the ceiling with using the handy type aneoeter. This conventional work has low productivity and high cost for safety anageent. Furtherore, this tendency is rearkale in roos with high ceilings such as entrance halls and factories. We propose a new easureent ethod using the UAV with air volue easuring device, and develop and evaluate the syste. The visual feedack control syste for self-localization is also developed ecause the GPS cannot e used indoors. In addition, a high ceiling roo with the diffuser is constructed as a ock-up for experient. We analysed the positioning errors of the UAV when hovering and when approaching and touching the diffuser. Also, we evaluated the touch perforance of the UAV to the ceiling y easuring the air volue. As a result, it was found that the UAV can ensure the perforance necessary for air volue easureent in this experient. Keywords UAV; Indoor Flight Technology; Building equipent works; Air Volue Measureent 1 Introduction In the Japanese construction industry, the nuer of young workers is decreasing and skilled workers are aging. For that reason, efforts to iprove productivity such as i-construction y using inforation technology (IT) and root technology are under way. One of the is the UAV (Unanned Aerial Vehicle). In civil engineering work, UAV are used for anageent of river facilities [1] and onitoring for ridge degradation [2]. In uilding work, quality is anaged y using iages taken fro the sky [3]. However, there are no cases of using UAV in uildings under construction. The ain cases are onitoring, confiration, and threediensional easureent y caeras and others. There are no cases of working in touch with the surrounding environent or eers such as transportation and installation. Building equipent works such as air conditioning, sanitation and electricity is ostly work in the uilding, especially in the vicinity of the ceiling using teporary scaffolding. In order to utilize UAV in uilding equipent works, it is necessary to perfor self-positioning without GPS, to recognize the surrounding environent, and to ake direct touch with the surrounding environent. One of the tasks near the ceiling in uilding equipent works is air volue easureent. Air volue easureent is a task of checking whether the aount of conditioned air lown out fro the diffuser provided on the ceiling surface of a roo is in accordance with the design value, and in general, total easureent is required. Conventionally, wind speed distriution iediately elow the diffuser is easured y a handy type aneoeter, and air volue is calculated y ultiplying the average value y the opening area. Although teporary scaffolds are necessary to approach the diffuser in a high place, work efficiency decreases due to assely, oveent and disassely of teporary scaffolds. Furtherore, safety anageent is iportant ecause it is work in a high place. The authors developed a device to easure the aount of air passing through the hood y ringing the hood for collecting wind into close touch with the ceiling surface y an elevating echanis [5]. As a result, the safety and productivity of the air volue easureent work have iproved. However, since
34 th International Syposiu on Autoation and Rootics in Construction (ISARC 2017) there is a liit to the height of the equipent, easureent using a teporary scaffold is required as in the past for diffuser in roos with high ceilings such as entrance halls and factories. Also, if ostacles such as furniture and production equipent are installed right under the diffuser, it is difficult to install the scaffold. On the other hand, if it is possile to easure the air volue using UAV, it can e expected that the prole when the person easures the air volue using the teporary scaffold or air volue easuring device as descried aove is solved. In this research, for the purpose of developing a UAV which can e used indoors in uilding equipent works, we developed a UAV capale of easuring air volue and evaluated its perforance. on UAV with caeras installed in the environent has the following two erits, and we chose this in the present study. 1. Since the target position can e set on the caera iage, it can e used, for exaple, when there is no 3D odel in the existing uilding or when the 3D odel does not confor to the current situation. 2 Syste Overview 2.1 Syste Configuration The new syste consists of UAV equipped with an Air Volue Measureent Unit, and a vision syste with caera and PC. The UAV is an oct-copter shown in Figure 1. In order to ount the Air Volue Measureent Unit in the center of the UAV, the frae was arranged in parallel crosses. The flight controller and the attery were arranged in a dispersed anner in the frae. The "Air Volue Measureent Unit" (Figure 2) consists of a Collect Wind Hood for collecting air flow fro the air inlet, a Sensor Hood with five wind speed sensors, and a control and counication circuit of the wind speed sensor and a attery. In addition, neoprene ruer was attached to the upper end of the Collect Wind Hood considering ease of adhesion with the ceiling surface. Furtherore, CFD (Coputational Fluid Dynaics) was perfored and the lower end of the sensor hood is set to e 90 lower than the propeller position, so that the airflow generated y the propeller did not influence the air flow easureent. Other specifications of the UAV, specifications of the caera and specifications of the wind speed sensor are shown in Tale 1. 2.2 Estiation of Position and Attitude using Vision Syste In this research, we decided to use a vision syste (Figure 3) that recognizes arkers attached to the UAV with caeras installed in the environent as a selfposition estiation ethod for indoor UAV that cannot use GPS. There are various ethods such as SLAM and inertial navigation as self-position estiation ethods in an environent where GPS cannot e used. We thought that the vision syste that recognizes arkers Figure 1. Appearance of UAV Figure 2. Air Volue Measureent Unit Tale 1. Syste specification UAV Flight controller Pixhawk Size [] D800 W800 H400 Frae aterial Caron fier Battery Lithiu polyer attery (2 pieces) Propeller size [in] 9.45 Weight [kg] 2.5 Sensor Acceleroeter (3 axes), gyroscope (3 axes), aroeter Caera frae rate [fps] 30 (Max) pixels 1920 1080 angle of view [ ] 90 wind speed sensor Measureent ethod Hot-wire Measureent range 0.05~12.0/s
34 th International Syposiu on Autoation and Rootics in Construction (ISARC 2017) 2. Since the easureent is perfored y the caera installed in the environent to e used, there is no need to install a sensor for recognizing the external environent in the UAV. As a result, the size of UAV can e iniized. Next, we descrie a ethod for estiating the position and attitude of UAV y recognizing arkers. The arkers on the UAV are two red circular arks (Figure 3). Two red circular arks are extracted fro the iage acquired y the caera and a circle is otained with the line connecting the centers of the two circular arks as diaeter, so as to find the area of the circle (A ) and the position on the screen (X,Y ). Based on the area of a circle (A ) otained fro two circular arks arranged 3 ahead (D ) so as to face the caera and the actual distance (X,Y ) to one pixel at 3 ahead, the horizontal coordinate of the caera (X) is given y Equation (1), the vertical coordinate of the caera (Y) is given y Equation (2), and the depth coordinate of the caera (Z) is given y Equation (3). Also, fro the size of the two arkers recognized y the caera, deterine the attitude of the UAV in the vertical axis rotation direction. The target position can e set y deterining the position and area of the circle on the caera iage. X X X A A (1) Y Y Y A A (2) Z 2.3 Control of UAV 2 D A A (3) Position control of the UAV is perfored as follows. The difference is otained fro the target position set on the caera iage and the UAV position, and the PID control as shown in Equation (4) is perfored so that the UAV position approaches the target position. In addition, when the deviation etween the target coordinates and the position coordinates of the UAV ecoes within the threshold value, it is autoatically controlled so as to aintain the height y using the aroeter. This was used as a safety easure to prevent the UAV fro falling when the UAV ecae uneasurale y the vision syste. Attitude control of the UAV is perfored as follows. The control ethod differs etween the rotation direction of the vertical axis and the two rotation directions of the horizontal axis. The rotation direction of the vertical axis perfors PID control using the direction of UAV with respect to the caera, and rotates the UAV so as to face the caera. The control of the two rotation directions of the horizontal axis is perfored autoatically y the flight controller so that the UAV ecoes horizontal. u(t) = K P d e(t) + KI e( )dt + KD e( t) dt (4) 3 Evaluation Experient In order to evaluate the function and perforance of the new UAV, we conducted a flight experient. The experiental site was a siulated ceiling with a diffuser as shown in Figure 4. In addition, we used the otion capture syste OptiTrack using an infrared caera to easure the flight path. First, we evaluated the asic perforance of the UAV y hovering. Furtherore, as the perforance required for air volue easureent, the approach perforance to the diffuser and the touch perforance with respect to the ceiling were evaluated. 3.1 Hovering Perforance 3.1.1 Experiental Method For confiration of hovering perforance, a target was set in the air, hovering for 30 seconds, and the flight trajectory was easured. In order to eliinate the disturance of flight caused y the UAV's own airflow Figure 3. Overview of vision syste Figure 4. Experient field (plan view)
34 th International Syposiu on Autoation and Rootics in Construction (ISARC 2017) during take-off, y passing the string through the center of the UAV, we restrained the UAV fro oving in the horizontal direction. After the UAV reached the target, the string was loosened and hovering was perfored. 3.1.2 Result Figure 5 shows the hovering flight trajectory (5 flights). The vertical axis is the vertical direction of the caera (Y), the horizontal axis is the horizontal direction of the caera (X) in the left figure, and the depth direction of the caera (Z) is shown in the right figure. The hovering perforance in the depth direction and the vertical direction is poor as copared with the horizontal direction. Furtherore, in order to confir the dispersion in hovering for each axial direction, Figure 6 shows the average value and the standard deviation of the difference fro the target position for each flight. When checking the axiu value, the average in the horizontal direction is 14, the standard deviation is ± 41, the average in the vertical direction is -49, the standard deviation is ± 119, and the average in the depth direction is -146, the standard deviation is ± 155. In the vertical direction, the standard deviation is large with respect to the horizontal direction. Also, in the depth direction, the dispersion of the average for each flight and the standard deviation is large with respect to the horizontal direction. 3.1.3 Discussion We will consider the reasons for the aove experient results. Factors related to the flight perforance of the UAV can e roughly divided into the following two groups. 1. Measureent perforance of the UAV y vision syste. 2. Control perforance y the flight controller. The difference etween the horizontal direction and the depth direction of the caera is in the easureent of the UAV y the vision syste. Generally, in threediensional easureent y the caera, the resolution of the caera depth is lower than the resolution of the caera plane. In this syste, the depth resolution is less than 1/10 of the plane. Also in the experiental result shown in FIG. 6, the average in the caera horizontal direction is 14, and the average in the caera depth direction is - 146. Therefore, it is considered that the easureent perforance of UAV y the vision syste is the cause of low hovering perforance in the caera depth direction The difference etween the horizontal direction and the vertical direction of the caera is controlled y a flight controller. The easureent perforance of the UAV y the vision syste is the sae resolution in oth the horizontal direction and the vertical direction of the caera. In the flight controller, when the control aount of the UAV y the vision syste is sall, it is autoatically controlled so that the height is aintained using the aroeter only in the vertical direction. In order to confir this control perforance, the output data of the aroeter was confired. Figure 7 shows the data converted fro the output data of the aroeter to the height difference (vertical axis) in tie series (horizontal axis). The standard deviation of the data is aout ± 200, which shows that the dispersion is large. Also, unlike the horizontal direction, the vertical direction is influenced y gravity, which akes control difficult. Therefore, low hovering perforance in the vertical direction is considered to e due to the control perforance y flight controller. Figure 5. Locus of hovering Figure 6. Difference etween easureent data and target point in hovering Figure 7. Height difference calculated fro the easureent data of the aroeter
34 th International Syposiu on Autoation and Rootics in Construction (ISARC 2017) 3.2 Perforance Required for Air Volue Measureent In order to accurately easure the air volue, it is necessary to accurately approach the UAV to the diffuser so that the diffuser enters the Collect Wind Hood. Therefore, the approach accuracy to the diffuser y the UAV was evaluated. Also, for accurate air volue easureent, it is necessary to collect all the conditioned air lown out fro the diffuser with the Collect Wind Hood, so the ceiling and Collect Wind Hood ust e in close touch with each other. According to the conventional ethod, a person is pushing the Collect Wind Hood against the ceiling, and the air volue is easured while aintaining that state. In order to check whether the state of touch with the ceiling of the UAV is aout the sae as the conventional ethod, the air volue was easured and evaluated. 3.2.1 Approach Perforance In the experient, the target position and the position to pass through were set, the flight was ade to fly fro the floor surface to the diffuser, and the flight trajectory was easured five ties. Approach to the target position was carried out after 25 seconds fro the start of flight. The flight trajectory is shown in Figure 8, and the difference fro the target position of the approach is shown in Figure 9. In Figure 8, the vertical axis is the vertical direction of the caera (Y), the horizontal axis is the horizontal direction of the caera (X) in the left figure, and the depth direction of the caera (Z) is shown in the right figure. The flight trajectory shows that the dispersion in the depth direction is large like the hovering perforance confired y the asic perforance. As for approach accuracy, the average is 25, the standard deviation is ± 15 in the horizontal direction of the caera, the average is 7, and the standard deviation is ± 263 in the depth direction of the caera, which shows that the dispersion in the depth direction of the caera is large. As discussed in the evaluation of asic perforance, we consider that the cause of large dispersion in the depth direction of the caera is the easureent perforance of the UAV y the vision syste with one caera. To iprove approach accuracy, it is necessary to construct a vision syste y using two caeras. The position of the UAV in the vertical direction and the horizontal direction of the caera is easured with one caera. Furtherore, another caera is installed y changing the position of 90 around the UAV fro the already installed caera. Then, the horizontal position of the UAV (the position in the depth direction of the other caera) is easured with the caera. By adopting the aove ethod, the position of the UAV can e otained without eing affected y the resolution of the vision syste in the depth direction of the caera, and the accuracy required for approach to the diffuser can e ensured. 3.2.2 Touch Perforance In the experient, we easured the air volue when the UAV pressed the Collect Wind Hood against the ceiling. As a conventional ethod, the person pressed the Collect Wind Hood against the ceiling, the air volue was easured, and the data was copared with the data easured y the UAV. In the air volue easureent, the wind speed was easured 5 ties at intervals of 1 second, the average wind speed was calculated, and the air volue was calculated. The easureent accuracy of the wind speed sensor is 5% of the easured data. Tale 2 shows the result of perforing air volue easureent three ties for oth the conventional ethod and the UAV. Copared with the conventional ethod, it was confired that the easureent result using UAV is within the easureent accuracy of the wind speed sensor. Fro this fact, we confired that touch with the ceiling of the UAV is coparale to the conventional ethod. Figure 8. Locus of flight
34 th International Syposiu on Autoation and Rootics in Construction (ISARC 2017) Figure 9. Difference etween easureent data and target point in approach position syste for structural inspection. Safety, Reliaility, Risk and Life-Cycle Perforance of Structures and Infrastructures, pages 281-288, 2014. [3] Nikkei Business Pulications. Photographed y UAV for construction record: Takenaka Corporation was full scale operation at Suita unicipal stadiu, NIKKEI CONSTRUCTION, 613, pages 25, 13/04/2015. (in Japanese) [4] K. Tanaka, S. Sakaoto and Y. Ae. Developent of an air volue easuring instruent: WINSPEC. In Proceedings of the 23rd ISARC, pages 210-214, Tokyo, Japan, 2006. Tale 2. Result of air volue easureent Conventional ethod UAV Measureent data [ 3 /h] No.1 380 No.2 363 No.3 346 No.1 371 No.2 350 No.3 354 Average [ 3 /h] 363 358 4 Conclusion For the purpose of developing UAV for use in indoor uilding equipent works, we ade a prototype UAV capale of easuring air volue and evaluated it. As a result, the approach accuracy to the diffuser is 25 on average in the caera horizontal direction and ± 15 in standard deviation. Fro this, y constructing a vision syste using two caeras, it was found that the accuracy required for the approach can e ensured at this experiental site. Also, fro the results of the air volue easureent, it was confired that touch with the ceiling of the UAV was aout the sae as the conventional ethod. In the future, in order to iprove the approach accuracy to the diffuser, we will construct a vision syste using two caeras and evaluate the flight perforance. References [1] S. Kuota and Y. Kawai. River aintenance anageent syste using three-diensional UAV data in JAPAN. ISPRS Annals of Photograetry, Reote Sensing & Spatial Inforation Sciences, pages 93-98, Issue 2, Vol.4, 2016. [2] S. Suitro, S. Nishiura, H. Matsuda and I. Bartoli. GPS-ased reote optical onitoring