Printer Friendly

Localization of compressed air leaks in industrial environments using service robots with ultrasonic microphones.

Abstract Compressed air is a widespread but costly energy carrier. Leaks account for 10 compressed air consumption in production facilities and their removal offers high potential for cost reduction. The turbulent flow from a leak causes broadband acoustic emissions. These are exploited for leak detection using a narrowband ultrasonic microphone that is insensitive to audible noise. A parabolic mirror or an acoustic horn is utilized to enhance the directivity and the received signal power of the microphone. The microphone is mounted on mobile service robots. In the project Robot}air{ an automated guided vehicle (AGV) and a remotely controlled micro aerial vehicle (MAV) were used for robotized inspection of production facilities. In order to detect leaks, predefined areas are scanned. A leak is detected and localized based on the sensed peak amplitude of the ultrasound signal. The corresponding pose of the sensor facing the leak is determined based on the selflocalization of the robot. Leak localization is carried out by triangulation of two sensor poses. Tests were conducted in an automobile production facility to evaluate the performance of the system. A leak was placed in an assembly line and multiple measurements were taken from two positions with the AGV and the MAV. The leaks were successfully detected and localized.

Keywords: air-coupled ultrasound, compressed air, leak detection, leak localization, mobile service robots

1. Introduction

Compressed air is used as an energy carrier in the manufacturing industry. The conversion from primary energy is costly and losses should be avoided. The losses due to leakage range from 10 project Robot}air{ [3] was to automate routine inspections and gain access to installations, which are otherwise hard to reach. This paper addresses the use of mobile robotic sensing systems for detecting, localizing and assessing compressed air leaks. An automated guided vehicle (AGV) with a scanning ultrasonic measuring unit is used for autonomous leak detection and localization in predefined areas in order to relieve factory workers of the monotonous task of manual leak search. Prior knowledge of the compressed air system and of common leak positions, such as couplings, fittings, joints, etc. [2], is used to narrow down the inspection targets in order to reduce the time demand for an inspection tour. A micro aerial vehicle (MAV) is used to inspect areas, which are hard or impossible to access by employees or by the AGV. It is operated manually, due to the often complex structure of the indoor airspace in industrial environments, e. g. hanging wires or narrow pipes.

2. Leak detection

Compressed air leaks can be detected manually with various methods. Leak detection spray is used to visualize the escaping air [2]. IR thermography can be used to visualize and detect the temperature gradient between the leaking component, which is cooled down by the decompression of the air, and its surroundings. References [4], [5] and [6] provide for a more detailed overview on approaches.

Acoustic emissions play a main role in compressed air leak detection [7]. The decompression of the compressed air causes turbulences near the leak. Pressure and density variations emerge from the turbulent flow as broadband acoustic emissions, which can be sensed from a distance using microphones. In industrial environments ultrasonic transducers are widely used to suppress noise in the audible frequency band. Own measurements showed little ultrasonic noise for most machinery except for presses, air blowers and active ultrasonic devices. For manual detection hand-held devices are used for leak detection [8].

In the recent years sensor arrays have been used for the detection of acoustic emissions of compressed air leaks. With microphone arrays the angle of arrival of a signal can be estimated without sensor movement. Signal processing techniques can also be used to separate the signal of the leak from interfering signals of other sources and additive noise. A handheld device with three ultrasonic microphones was used to estimate the direction of arrival using the time difference of arrival [9]. Simulations for an array with 32 randomly placed ultrasonic microphones were conducted in a scenario with multiple leaks [10]. An acoustic camera, a combination of a microphone array and a camera, was used to localize and visualize compressed air leaks [11].

Compressed air leak detection with mobile robots is rarely addressed in literature. A similar leak detection system using an AGV and a single directed sensor for remote gas detection and localization was addressed in [6, 12]. In [6] a scanning method ("dynamic raster scan") is used to optimize scanning time and coverage of the scanned area. In [12] a combination of triangulation and maximum method is used to reposition the AGV to improve the accuracy and precision of the estimated leak position. The intelligent automatic measuring system used for the tests in this paper was investigated under laboratory conditions in [13]. Methods for quick scanning, leak detection and estimation of the direction of the leak were introduced. The paper showed the feasibility of using the system and discussed boundary conditions, which influence the accuracy of the methods.

3. Robotic system

The robotic system consists of an AGV, a MAV (Figure 1) and a movable control station. All subsystems communicate using 5 GHz WLAN. They are controlled using the ROS (Robot Operating System) framework.

3.1 Automated guided vehicle

The AGV is a vehicle custom made by S-Elektronik GmbH & Co. KG in cooperation with Fraunhofer Institute for Communication, Information Processing and Ergonomics. It is maneuvered by two diagonal center pivot plate drives. 2D mapping and self-localization are realized by diagonally mounted laser scanners SICK S300. On top of the rear side of the AGV is a landing platform for the MAV. A stereo camera system on a pan-tilt-unit (PTU) Schunk PW 70, which is used to track the position of the MAV, is mounted in the middle top side of the AGV. An intelligent measuring system is mounted in the front of the AGV (Figure 2, left). It consists of a measurement device carrier with several sensors, a PTU and a computer. The sensor of interest in this contribution is a piezoelectric microphone with a parabolic mirror produced by SONOTEC Ultraschallsensorik Halle GmbH to focus the incoming ultrasonic wave towards the sensor. This results in an increased range and a higher directivity of the sensor. The microphone operates in a narrow frequency band with a resonance frequency of 40 kHz. The ultrasonic signal is converted to an audible signal by a frequency mixer, which is recorded with 44100 samples per second and 16 bit resolution using an uncalibrated sound card. The recorded signal is proportional to the incoming sound pressure.

A camera is mounted behind the parabolic mirror for documentation purposes. The sensors are placed on the measurement device carrier, which is mounted on a PTU Schunk PW 90. Its two axes have a positioning reproducibility of 0.004. The PTU enables scanning target areas and directing the sensors independent of the AGV. The geometric transformations between the local coordinate systems of the AGV, the PTU and the sensor result directly from the technical design of the devices and the mechanical assembly, and were calibrated manually.



3.2 Micro aerial vehicle

The MAV is an Aibot X6 multicopter (Figure 2, right) made by Aibotix GmbH. The MAV was equipped with six colored LED markers for tracking with the stereo camera system of the AGV. The payload for compressed air leak detection consists of an ultrasound microphone, a camera and a light-weight computer. Since turbulence originating from the MAV's rotors is emitting noise similar to compressed air leaks, an acoustic horn is used to shield the microphone, a model similar to the one used on the AGV, against this noise. The sensor is placed on the rim of the frame facing away from the rotor turbulence to reduce the influence of the noise further. The signal acquisition is similar to the one used for the AGV. The camera points to the same the measuring direction as the microphone. The sensor pose was measured manually with respect to the pose of the MAV.

4. Leak Detection and Localization

4.1 Automated inspection with AGV

For the AGV an automated leak localization method was developed. In the beginning the inspection route is taught-in manually using the inspector's knowledge of the compressed air network and of fixed and temporary obstacles in the inspection area. Measurement positions are chosen and a measurement area is defined using spherical coordinates with the PTU as origin. The route is automatically planned and optimized for short, collision-free paths. The inspection area is scanned line by line from each specified measurement point. The level of the ultrasonic signal is recorded as a function of the PTU pan angle and the maximum is calculated for each line. If the detection criterion, a sufficient ratio between the maximum and the offset of the signal, is met a vertical scan is performed that crosses the found maxima. More details on the scanning method and the detection criterion are given in [13]. If a leak is detected the sensor is directed towards the leak and its pose is saved for localization. A picture of the leak augmented by the measurement direction and the ultrasound level of the leak is saved for documentation.


When a leak was detected from multiple positions, the leak localization is performed. The leak location is calculated using triangulation. Two sensor poses are interpreted as rays in the three-dimensional space, with their intersection representing the position of the leak. Due to estimation errors regarding the position of the AGV and the pose of the PTU a direct intersection of both rays is unlikely. Therefore, the points on both rays that have the smallest distance from each other are determined. The point in the center of the line segment between those two points is the estimated leak position. When the origins of two sensor poses are close to each other or their directions are nearly parallel, no localization will be attempted, to avoid scenarios which are prone to large estimation errors. An example of a successful leak localization from two measurement positions is shown in Figure 3 in a 2D map.

4.2 Piloted inspection with MAV

The MAV is operated manually. Due to the continuous movement of the MAV, a complete scan of an inspection area analog to section 4.1 is not possible. Therefore, the operator manually directs the MAV close towards the target and the potential leak positions. To support the operator several assistance functions are provided. The audio signal of the leak is transmitted to headphones, such that the operator can perform an aural inspection analog to scanning methods using hand-held devices. Also a live video broadcast can be streamed to smart glasses or to a monitor, enabling the operator to align the sensor more precisely based on the visual cues. To account for smaller variations in the MAV pose and the transmission delay of the audio/video data, a short recording can be triggered while aiming at a potential leak. When the recording is finished the sound level maximum is automatically searched within the recording. The corresponding picture and the sensor pose are documented. The leak localization is performed analogous to the previous section.

5. Industrial case study set-up

An assembly line for car transmissions was used for the case study. Since there were no real leaks present due to recent maintenance, a steel pipe, 21 mm in diameter, with a round drilling of 1 mm diameter was used as an artificial leak. The pipe was mounted on top of a machine in the assembly line (Figure 4) and connected to the common compressed air supply with a pressure of approximately 8 bar.


The world coordinate system for leak localization is a right hand Cartesian coordinate system based on the 2D map recorded beforehand by the AGV. The position of the leak was approximated by measuring the distance from landmarks in the 2D map and the height of the leak. Small errors resulting from self-localization, tracking of the MAV, and geometric calibration of the sensors are likely to produce biased results for the estimated leak locations. Therefore, the focus of the tests lies on the repeatability, i. e. the standard deviation, of the estimates. The results of the detections were validated manually by a visual inspection of the result pictures with the estimated leak position (Figure 4).

The AGV was positioned in front of the leak at two different positions (Figure 5). 30 scans were conducted from each position without moving the AGV between the scans to reduce stochastic errors from the self-localization. For each scan the sensor pose, the pan angle [empty set] and the tilt angle [theta] are recorded. For both angles the mean and the standard deviation are calculated. As a measure of overall precision of the estimated direction the combined standard deviation [s.sub.[empty set],[theta]] of the angle pairs (1) is calculated [13]. N is the number of successful detections and i is the index of the detections.


The results for the estimated leak positions are treated likewise. The mean and standard deviation of the three Cartesian coordinates x, y and z are calculated and the combined standard deviation [s.sub.x,y,z] is calculated:


[N.sub.P] is the number of estimated positions and k is the index of the positions.


The MAV was operated manually to face the leak from some distance (Figure 6). Using the onboard camera and the visual aid, the operator initiated multiple detection attempts from random positions in front of the leak. The copter poses and the pictures of the leak were saved. Two data sets were used to calculate the results. On the one hand all detections were used. On the other hand only detections where the leak was placed within the marking in the result picture were used. The combined standard deviation [s.sub.x,y,z] (2) calculated analogous to the AGV-based tests.


6. Results

In case of the AGV, for both positions all detections were successful and passed the visual inspection (N = 30). The results of the direction estimation are shown in Table 1. The combined standard deviation [s.sub.[empty set],[theta]] is in the same range as in a similar scenario in a laboratory environment [13]. The standard deviation of the tilt angle [theta] is about twice the size than the one for the pan angle .

Using all pairs of sensor poses from the detections 30x30 locations ([N.sub.p] = 900) were estimated (Table 2). The spatial distribution of the positions is visualized in Figure 7.


In case of the MAV, nine detection attempts by the MAV were successfully carried out and recorded. Three of the detections were rejected after the visual inspection. Due to their random nature not every pair of sensor poses was suited for a successful localization. The results of the measurements are summarized in Table 3. The spatial distribution of the estimated leak positions is visualized in Figure 8. Outliers can be seen in the visualization of both data sets.


A comparison of the approximated leak position and the estimated leak positions for all data sets is shown in Table 4. The Euclidean distance d is calculated in respect to the approximated leak position.

7. Discussion

The results for leak localization with the AGV look very promising. The standard deviation of both the direction estimation ([s.sub.[empty set],[theta]] < 0.3[degrees]) and the location estimation ([s.sub.x,y,z]<= 0.03 m) are very low. In comparison, the location estimates of the leak using the MAV have a much higher standard deviation ([s.sub.x,y,z] > 1 m) although most result pictures suggest that the sensor poses point towards the leak. On the one hand the data sample from the MAV is much smaller, such that single outliers carry a much larger weight on the result. On the other hand the estimation of the sensor pose is less precise due to an additional error source: The position estimation using the AGV is mainly influenced by the self-localization of the AGV while staying stationary, the geometric calibration of the sensor, and the precision of PTU and ultrasonic sensor. The first two bias the leak position and only the latter influence the standard deviation. The MAV is tracked while moving and so another influence on the standard deviation is added. A more precise method to localize the MAV is expected to greatly improve the leak localization results. The difference of the standard deviations of the estimated PTU angles can be a result of the geometry of the parabolic mirror, which is not rotationally symmetric. Since the vertical detection is based on the angle of the horizontal detection the stochastic error of the second part of the detection can increase. Although, calculating the standard deviation of the angles from the data of the laboratory experiments [13] for comparison cannot support either hypothesis. The vertical structures in Figure 7 are a result of using each direction estimate multiple times. Therefore, the location estimates using all direction estimates from one position and a single directions estimate from the other position will form such a distribution located along the ray from the latter direction estimate. The results for the leak localization using the MAV do not improve by manually discarding incorrect detections. As seen in Figure 8 both distributions have a similar spread. Since the corrected data set is much smaller than the original one, the influence of the spread on the standard deviation is larger.

8. Summary and outlook

In the paper it was shown that mobile service robots can be successfully used for localization of compressed air leaks in industrial environments. Due to the high precision the AGV is suited for autonomous leak inspection, where the estimated leak position can be integrated into a 3D map of the inspection area. In addition to automatically triggered maintenance the data can be analyzed for maintenance scheduling. The MAV can be used as a leak detection tool for areas that are hard to access by foot, but offer a sufficient flight corridor. A rough estimate of the sensor pose and the picture of the leak suffice for manual maintenance. Since improvements in indoor self-localization and autonomous flight are under way, it can be expected that a MAV can be used for autonomous leak inspection with a high maneuverability in the future.


The authors would like to thank the Federal Ministry of Education and Research (BMBF) for supporting the project Robot}air{ (reference 01IM12007F) under which this work has been carried out. The contents of this publication are the sole responsibility of the authors.


[1.] P Radgen and E Blaustein, 'Compressed Air Systems in the European Union', Stuttgart: LOG_X Verlag, 2001.

[2.] C Galitsky and E Worrell, 'Energy Efficiency Improvement and Cost Saving Opportunities for the Vehicle Assembly Industry', in LBNL-50939-Revision, March 2008.

[3.] J Hegenberg, R Herrmann, D Ziegner, L Schmidt, T Guenther, A O Muller, A Kroll, T Barz and D Schulz, 'Forschungsprojekt RobotAir: Praxistaugliches Boden-LuftServicerobotersystem fur die Inspektion industrieller Druckluftversorgung und die Verbesserung der Arbeitsumgebungsfaktoren', in Technische Sicherheit, Vol 5, No 5, pp. 16-22, May 2015.

[4.] A Kroll, W Baetz and D Peretzki, 'On autonomous detection of pressured air and gas leaks using passive IR-thermography for mobile robot application', in IEEE International Conference on Robotics and Automation (ICRA 2009), pp. 921-926, Kobe, Japan, May 2009.

[5.] W Baetz, A Kroll and S Soldan, 'On gas leak detection of pressurised components by using thermograms and pattern recognition algorithms', in 8th Int. Conf. on NDE in Relation to Structural Integrity for Nuclear and Pressurised Components, pp. 503-512, Berlin, Germany: European Commission, Joint Research Center, Institute for Energy, 29. September - 1. Oktober 2010.

[6.] W Baetz, A Kroll and G Bonow, 'Mobile Robots with Active IR-Optical Sensing for Remote Gas Detection and Source Localization', in IEEE International Conference on Robotics and Automation (ICRA 2009), pp. 2773-2778, Kobe, Japan, 2009.

[7.] A Pollock and S-Y Hsu, 'Leak detection using acoustic emission', Journal of Acoustic Emission, vol. 1, pp. 237-243, 1982.

[8.] Compressed Air Leak Survey Handbook. Brussel: SDT International, 2009.

[9.] X Li, H Liu and X Yang, 'Sound source localization for mobile robot based on time difference feature and space grid matching', in IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 2879-2886, San Francisco, CA, September 2011.

[10.] J Steckel and H Peremans, 'Ultrasound-based air leak detection using a random microphone array and sparse representations', in IEEE SENSORS 2014, pp. 1026-1029, Valencia, Spain, November 2014.

[11.] P Eret and C Meskell, 'Microphone arrays as a leakage detection tool in industrial compressed air systems', Advances in Acoustics and Vibration, pp. 1-10, 2012.

[12.] G Bonow and A Kroll, 'Gas leak localization in industrial environments using a TDLAS-based remote gas sensor and autonomous mobile robot with the Tri-Max method', in IEEE International Conference on Robotics and Automation (ICRA), Karlsruhe, Germany, 2013.

[13.] T Guenther and A Kroll, 'Automated detection of compressed air leaks using a scanning ultrasonic sensor system', in 2016 IEEE Sensors Applications Symposium (SAS), pp. 116-121, Catania, Italy, April 2016.

Thomas GUENTHER (1), Andreas KROLL (1)

(1) Department of Measurement and Control, University of Kassel; Kassel, Germany

Phone: +49 561 804 2758, Fax: +49 561 804 2847; e-mail:,
Table 1. Results of leak direction estimation from two positions

Data set    N   [empty set] in  [s.sub.[empty set]] in  [theta] in

Position 1  30  -16.95                  0.12            -17.36
Position 2  30   -3.70                  0.13            -13.95

Data set    [s.sub.[theta]] in  [s.sub.[empty set],[theta]] in

Position 1        0.26                     0.29
Position 2        0.24                     0.27

Table 2. Results of leak position estimation for the AGV

Data set  [N.sub.P]  x in m  [s.sub.x] in m  y in m  [s.sub.y] in m

full         900      18.64    0.01           20.63   0.02

Data set  z in m  [s.sub.z] in m  [s.sub.x,y,z] in m

full       2.35       0.02            0.03

Table 3. Results of leak position estimation for the MAV

Data set   [N.sub.P]  x in m  [s.sub.x] in m  y in m  [s.sub.y] in m

full       18         18.87        0.80       21.03        1.15
corrected  6          19.36        1.15       20.51        1.64

Data set   z in m  [s.sub.z] in m  [s.sub.x,y,z] in m

full        2.29        0.17              1.45
corrected   2.40        0.25              2.21

Table 4. Comparison of approximated leak position and the mean of the
estimated leak positions

Data set         x in m  y in m  z in m  d in m

approximation    19.02   20.77   2.20    0.00
AGV - full       18.64   20.63   2.35    0.43
MAV - full       18.87   21.03   2.29    0.31
MAV - corrected  19.36   20.51   2.40    0.47
COPYRIGHT 2016 Acoustic Emission Group
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2016 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Guenther, Thomas; Kroll, Andreas
Publication:Journal of Acoustic Emission
Article Type:Report
Date:Jan 1, 2016
Previous Article:Measurement of electromagnetic emission as a tool to study fracture processes of carbon fibre reinforced polymers.
Next Article:Modeled transverse surface ae signals from buried dipole sources in a polymer rod.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters