Computer Vision Application in Automatic Meter
Calibration
Leo, S.L. Pang Technical Services Department CLP Power Hong Kong Limited
Hong Kong
slpang@clp.hk
Dr. W.L. Chan  Department of Electrical Engineering  The Hong Kong Polytechnic University
Hong Kong
eewlchan@polyu.edu.hk
Abstract—CLP Power Hong Kong Limited (CLP Power) is a leading electricity utility in Hong Kong. Instrument calibration is an important supporting service for the company. Under traditional practice, manual calibration is being performed. As one of the new performance improvement initiatives, computer-vision application in meter calibration was introduced in 2004. This paper summarised the appl
ication of an intelligent automatic calibration system for instruments without computer interface output. This system is designed to communicate with the high precision calibrator using General Purpose Interface Bus (GPIB) interface to give out pre-programmed test signals and connect to a camera through the PXI-1490 interface card to capture meter readings in image format and then convert to digital format for test data verification in computer automatically. Now, the company is enjoying the benefits from the system in terms of eliminated human calibration errors and reduced operating costs.
Keywords- computer vision, automatic, meter, calibration
I.I NTRODUCTION
CLP Power, being the largest power company in Hong Kong, is supplying electricity power to over two million customers in its supply areas.  As one of the success factors, CLP Power’s operation and maintenance teams need to be equipped with different kinds of portable instruments enabling them to be the most efficient teams handling daily works for the power system. As such, portable instrument periodic calibration is regarded as part of the important technical support service for the company. To suit the business needs, there are various types of instruments being used and more than 3,000 units are handheld with either digital or analogue displays.  The Technical Services Department’s electrical s
tandard laboratory, as the company service provider, is responsible for providing the calibration service in order to ensure the accuracy and functionality of these instruments. Prior to establishing the proposed system, the tests were performed manually by controlling the calibrator to output a series of test signals to the instruments under calibration, reading out and logging the measured values, comparing with the allowable limits and transferring the handwritten raw data to computer.  This manual process required long calibration time (about one hour per unit on average) and had to be tested by a competent operator. Moreover, the process was also susceptible to human errors in which the department was most concerned about.  In view of assuring work quality and reducing cost, CLP Power therefore initiated an innovative idea to tailor-make a new automated calibration system by computer vision technology. Though computer-based automatic calibration system is not new [1], digital interface is required in order to collect instrument readings by the computer. As one of the computer vision technology advantages, the requirement of a separated digital interface is no longer needed and calibration system will provide a very low cost solution for low cost instruments without computer interface. Computer vision is a maturing technology and it has been used in many areas [2]. In this paper, an automatic instrument calibration system using computer vision is reported.
II.MANAGING  INSTRUMENT IN CLP POWER CLP Power as a world class electricity utility, the suppl
y reliability performance is over 99.99%. Instrumentation calibration is essential to supporting this excellent performance. Currently, there are about 6,500 units of portable instruments divided into 28 classes that are being used in daily activities. All these instruments require periodic calibration according to the company’s quality management system requirements. In this connection, the department has to provide about 3,500 number of instrument calibration services yearly including calibration after repair and new instrument acceptance test.
Such a large calibration volume consumes lots of manpower. To ensure high quality calibration services are delivered to meet the service pledges, different effective strategies have been explored, including:
i.Obtaining optimal calibration intervals for different
classes of instruments by experience and statistical
method
ii.Developing virtual instruments to reduce number of traditional instruments to be used hence minimizing
calibration efforts; and
iii.Automating calibration process [4].
It was revealed from the analysis that  7 major classes of instruments, digital multi-meter, analogue multi-meters,
clip-on meter, insulation resistance meter, power factor & phase angle meter,  loop impedance & earth tester and residual current device represent more than 40% of the 28 instrument classes with a total  quantity of about  2,800 units. Should all these types of instrument calibration be automated, about 30% of manpower saving in calibration can be obtained yearly. More importantly, most of the possible human errors can be eliminated compared with the existing manual calibration practice.
III.S YSTEM O PERATION P RINCIPLE
The automated test station implemented by CLP Power is based on machine vision technology. The system consists of an industrial computer with PXI (PCI eXtensions for Instrumentation) interface and a PXI-1409 which is a high performance machine vision hardware used for capturing the images of the portable instruments under test. The control software used to process the images was developed base
d on the vision development module which is supplied by the hardware manufacturer.The hardware connection is shown in Figure 1 and the automated test station goes through the following steps:
1)By using the GPIB interface of a traditional calibrator, the PXI system automates the calibrator to sequentially output the specified test signals as programmed according to the existing test forms.
2)The PXI-1409 image card acquires the images of instrument reading displays.  By identifying a reference edge and regions of interest (ROI), the test system converts the numerical or pointer deflection values as shown in the digital and analogue meter displays to digital formats.  The data is then logged to a file.
3)Steps 1 and 2 repeat until all test points are tested.
4)Software programmatically compares the results with the allowable limits, and then generates “pass” or “fail” test results accordingly.  This pass/fail data are also logged to file.
Figure 1 - Hardware Connection Diagram
IV.M ECHANICAL A RRANGEMENT OF THE
C AMERA
A.Mounting Assembly
To allow the system capturing correct image signal from the meter under test, the image positions have to be pre-defined. First of all, the unit under test has to be fixed to a relative position under the camera without any rotational and lateral shifts. A tailor-made mounting assembly for serving the purpose is therefore treated as the pre-requisite requirement for capturing steady image.
B.Reference Point for ROI
Having the mounting assembly fixed, next steps are to define the reading image reference point and the  locations of the meter characters or numerical readings to be displayed on the meter screen, so called region of interest. Different types of meters will have different ROI. Locating a reference framework is vital in the system as the ROI will be relative to the framework. The system is using software to locate a reference coordinates as the reference points and the results are shown in Figure 2 below.
C.Lighting Control
As the lighting of the environment will affect the grey level value of the digitized image, it is important to have a constant lighting environment so as to make the recognition process to be stable. Equipment background light if built-in may require to be turned on during calibration.values翻译
Figure 2 – Reference Point for ROI of Different DMs
V.D IGITAL METER
Before performing meter calibration of a digital meter (DM), all ROI have to be set so as to allow the system to know what to be recognized from the display of a DM as shown in Figure 3. In CLP Power Electrical Standard Laboratory, the ROI of three DMs were defined. The different regions in the display will display different readings like Voltage, Current, AC, DC, etc.
For a reading, say 0.2 mVdc, the regions of ‘0.2,
DC’,‘V’,’m’ will be displayed. The system will then check for
C.A. 5003 BM222
Fluke 189 Fluke 87
the grey level of displayed ROI. Since the blank region will have much higher grey level values than those regions with characters, images were distinguished and read by the system from this great grey level difference. For the digits, the regions of the numerical digits were mapped with the internal 7-segment library to produce the closest digit with the one shown in the display.
Figure 3 – Defined ROI a DM
VI.ANALOGUE METER
Unlike DM, analogue meter (AM) is, to make use of a pointer deflection to indicate measured values like CA5003, the ROI is the two red lines from zero reference to the expected full scale deflection as shown in Figure 4. They are in fact the defined ROI for this type of meter different from DM. When signal is fed into the meter, the meter pointer will deflect to the corresponding position within the two red lines. The readings of the meter will then be calculated from the portion ratio of the pointer inside the two red lines multiplied by the total range.
Figure 4  – Defined ROI of a AM
VII.S OFTWARE F EATURES
The software of the automatic calibration system is running under Windows XP Professional operating system platform and it is written in image capturing application program environment. The features of the software are listed as follows:
•Verification of authorized users
•Interface with barcode reader for scanning operator ID and meter ID
•Mapping operator and meter information from database
•Interface with calibrators to output programmable signals
•Selection of correct calibrators
•Calibration of the defined types meters
•Selection of  one or two meters to be calibrated at the same time
•Selection of calibration modes, single point or continuous point calibration
•Test results logging and verifying
•Test certificate and report printing
VIII.SMART FEATURES INCORPORATED To make the system more successful, suitable smart features are recommended to be built-in. First consideration is to remind the operator whenever a wrong test item is selected during the calibration process. That is, when the operator inadvertently selected a wrong test item the system will not continue the calibration and will prompt the operator to select the correct one before giving out calibration signal to protect the test unit. Secondly, some sorts of automatic image searching feature is to be built-in to allow operator to take out the test unit for inspection any time during calibration. With this smart feature the system is able to recall the remembered reference point and ROI even if the test unit is taken out and then put back after some time to continue the calibration.
IX.REPORTING SYSTEM
To allow operators to view the test results as shown in Figure 5 before producing test report, the reporting system is designed to let operators highlight any particular row of result for re-testing if encountering any questionable results . Once all the test results were verified, the system will allow op
erators to choose ‘Print Report’ or ‘Print Certificate’ as required. After that, the ‘Open spreadsheet’ program will be invoked to show the test report or certificate. Users can then modify the format, and add information to the file before actually printing it out. Should an electronic test report be required, the test report generated can be saved to the report database directly without scanning efforts.
Figure 5 -  Test Report Generating During Calibration
X.S UGGESTION FOR SOLVING THE FIXED
MOUNTING PROBLEM
For analogue meter, it is possible to solve the fixed mounting requirement by using computer vision algorithm [4]. During initialization, the image of a meter with zero reading is retrieved.  Another image of the same meter captured by the same camera at the same location under the same illumination will be grabbed.  As shown in Figure 6, a "maximum" operation is carried out between the two images, resulting in an image of the meter without the pointer or the bar.  This image is treated as a reference and is stored inside the memory of the computer.  During normal operation, a real-time image of the meter is grabbed and an "image subtraction" algorithm is then executed to compare this real-time image and the reference image stored in the computer memory.  The operation is shown in Figure 7. H
ence, a pointer is then highlighted as shown in Figure 8.  "Hough Transform" is then carried out to locate the most probable position of the pointer and those noisy points can be eliminated.  A "robust line fit" algorithm is employed to identify the optimal position and orientation of the pointer.  Once the orientation is known, the real-time reading can be obtained by referring to the table within the database kept in the memory of the computer.  The time-stamped value is then stored for trend logging and statistical analysis.  The flowchart of the method is shown in Figure 9.
Figure 6  – An Illustration of the "Maximum" Operation
For Hough Transform, a straight line can always be written as:
x cos θ + y sin θ = r
where θ is the angle that a normal to line makes with the x-axis and r is the length of this normal.  Any point, (xi, yi) on this line must satisfy the straight line equation while the equation can be interpreted as a sinusoidal curve in the (r, θ) space.  If a number of points are collinear in the (x, y) space, their Hough transform curves must intersect in the same point in the (r, θ) space.  Each point (xi, yi) will contribute a count to the cells given by the equation and the cells with high counts will give the desired lines.  The value θ gives an approximate position of the meter pointer.  To refine the meter reading, the optimal line that represents the point can be obtained by means of robust fit of the useful pixels.  The line fitting is based on minimization of the squared deviation of all useful pixels from the optimal line which has the equation "y = a + bx".
()
()x b-a-y x
=
;
1
x
b-
y
=
a
x
b-a-
y
i
i
i
i
i
i
i
∑
∑
∑
∑
Min
Figure 7  – An Illustration of the Subtraction Operator
Figure 8  – The Extracted Meter Pointer
Figure 9  – The Flowchart of the Method
XI.T ANGIBLE B ENEFITS
With the automatic calibration system as shown in Figure 10 in place, the test time has been decreased significantly from approximately 1 hour to less than 10 minutes per unit, i.e. 6 times of speed improvement for a typical DM out of the four developed meter models as shown in Figure 11. Apart from time saving, instrument calibration by competent operator becomes less demanding as the process has been fully automated. More importantly, possible human errors are eliminated. Since the complexity of the task has been simplified, training for operator on manual calibration can be much simpler.
Figure 10 – Two DM Being Calibrated By The System
Figure 11 – Four Types of Meters Developed Automatic Calibration Program By Computer Vision
XII.C ONCLUSION
With the help of advanced computer vision technology, automating calibration and testing for instruments without computer interface is technically feasible and cost effective for any utility which needs to manage large number of same model instruments with digital or analogue display. Such system not only saves manpower but also improves the quality of work. Apart from computer technology, other applications such as automatic calibration by infra-red signal and virtual instrument development are also worth considering  in terms of work quality improvement and cost saving. This newly developed computer vision system could also benefit other industries.
References:
[1]S.C. Wang, C.L. Chen, “Computer-aided transducer calibration system
for a practical power system”, IEE Proc. – Sci. Meas. Technol., Vol.142, No.6, Nov. 1995, pp. 459-462
[2]  A.T.P. So, W.L. Chan, K.C. Li, "A Computer Vision Based Fire
Detection, Lighting and Air-conditioning Control System", Proceedings of CIBSE National Conference, Computers in Construction Industry, Manchester, May 1993, pp. 345-355
[3]W.L. Chan, Leo S.L. Pang, C.F. Ma, "Computer Vision Applications in
Power Substations", Proceedings of the 2004 IEEE International Conference on Electric Utility Deregulation, Restructuring and Power Technologies, DRPT 2004, Hong Kong, April, 2004, pp. 383-388
[4]Leo S.L. Pang, " The First Successful Application of Computer Vision
Technology in Automating Multi-meter Calibration” CLP Power TSD Technical Bulletin Issue No. 6, Aug 2004.

版权声明：本站内容均来自互联网，仅供演示用，请勿用于商业和其他非法用途。如果侵犯了您的权益请与我们联系QQ:729038198，我们将在24小时内删除。