GB2556296A - Method for mapping fuel assemblies - Google Patents

Abstract

The invention relates to the control of positions of fuel assemblies arranged within a nuclear reactor vessel, each fuel assembly including at least one reference element. The following steps are carried out: acquiring images taken by a camera filming at least the reference elements of at least one portion of the assemblies; denoising the acquired images, -image processing to locate the reference elements in the denoised images. More particularly, the camera (CAM) films in close-up only one portion of the assemblies in each image, and moves along a predetermined path to obtain a sequence comprising consecutive images of all the assemblies. The steps of image denoising and processing are then carried out in at least one area of interest defined in each acquired image, the area of interest including one or more expected reference element positions, with a tolerance around said expected positions to also include real positions of the reference elements in the acquired images, said area of interest being predetermined in each image as a function of said predetermined path.

Description

(56) Documents Cited:

US 5594764 A (58) Field of Search:

INT CL G21C Other: EPO-Internal

US 20130195237 A1 (87) International Publication Data:

WO2017/037384 Fr 09.03.2017 (71) Applicant(s):

Electricite De France (Incorporated in France)

22-30 avenue de Wagram, 75008 Paris,

France (including Overseas Departments and Territori es) (72) Inventor(s):

Vincent Le Guen Denis Vautrin Nicolas Paul Pierre-Louis Filiot (74) Agent and/or Address for Service:

Mathys & Squire LLP

The Shard, 32 London Bridge Street, LONDON, SE1 9SG, United Kingdom (54) Title of the Invention: Method for mapping fuel assemblies Abstract Title: Method for mapping fuel assemblies (57) The invention relates to the control of positions of fuel assemblies arranged within a nuclear reactor vessel, each fuel assembly including at least one reference element. The following steps are carried out: acquiring images taken by a camera filming at least the reference elements of at least one portion of the assemblies; denoising the acquired images, -image processing to locate the reference elements in the denoised images. More particularly, the camera (CAM) films in close-up only one portion of the assemblies in each image, and moves along a predetermined path to obtain a sequence comprising consecutive images of all the assemblies. The steps of image denoising and processing are then carried out in at least one area of interest defined in each acquired image, the area of interest including one or more expected reference element positions, with a tolerance around said expected positions to also include real positions of the reference elements in the acquired images, said area of interest being predetermined in each image as a function of said predetermined path.

WO 2017/037384

PCT/FR2016/052157

1/13

FIG.

TRS <%AAA .......*<, •-Λ-λΛα'.ΑΑΧΧΑΑΑΑ^ AA•ί^'/ίί'ζΛιΛχΑ^ΑχΧ’^χΧΧ^ΧΧ: AaaAA ViW&^\* · A vNaA AAAAA

Α^ΑΑΑΑΑ-ΛΛίίΛΛ·«Ά.

J

A^AA^AAXxX<XXX^,^jxX:*X'|>Z^>'i^5i:

A ,ΑΑΑΑ^ΪαΑΑAAAAAaAjAV.Vi.·' .'.· . &.'.

'-S^^7\^sG5^x'^4Z>^>>^AAAAAA;\\j^^\,Ay.ZAA\£z.

...........Α^ΑΑΑΑΑΑχΆ^\ΑΛΛΛ.'*.*ζΑίίν'ΛΑν»Α\'ΐ^\ΑΑΑΑΑΑΑίΪΑΑΛΑΛΑ’·

AVZ>^^k-A^»Ai^<^^AAA^AA'AA^7^fcAAA^*>_rV.-^*^-’**iA.AA^A^ji_%\A^A>AS^AA>^>y'.'4:i$':*AX^ •ζΑΑΑ A A ί ^A* AAAAA'S^A A Α-'Ά MA^t AAAAAAAA^^A A Λ»

..A AAA A a^KaAAAAAAA^AVVAA.' :( : : 5 : : :: v :: * if ·. 4 't . aAAAAAAAAA^^AAAA*

S .. ? Ϋ ...... :

-aaaaaaA^.a.aXaaaj.a¹ > 4- aN AAA ''S' ❖’^sS&a>a^xa,4^A\a\^a<W ✓> j . .. .. .. ,Ϊ\αΑΑΑΑ\ΑΛαΑα. - ...1¼..........

aV A & . . *t\ A AAAAA *^A> , ί '£.·. AA ΛΑ A?W\AA<A.AAAAf)A*/Z,

A^.A\AA X·*·^ - ΆΑΑΑ^ααΑ,ΑιΑ.^ι -* ,χ.χ^χΑΑ^^\Α,Α^Α\^.ν^^*:^·^Α.ΑΑΑ^\\Α^!\Α\Α·^.>ΑΑ^χ-·-;-·^«χ<<ΑχΑχΑ5\\\>^.Α>\Α5ΑΑν<ί^<^ :/4/: :J J :< j: < $ > : : : :]· <

Λ·ΛΜ.·ΛΑΑΑΑΑΑ^ι_νΑ ΑΑΑΑΧΧΧ·^.·.·,·.-,-ΛΛ·»'ίί> — - · A AA^AAAAAAAA^AAVaV, Z-> X^.· a . ·α . a . ^χαΑΑΑΑΑαΛ .ΑΑίΑΑΑ5ί\ΑΑΑ§.ΑΑ?<-.'νΖ,Α;ν.^^<νΑΑ\.ΑΑΑΝ<ΑΑΑΑ,·. '^ΑΑΑ·.ν,ΑΑ^ . ,..γ_<,„·,ΑΑΑΑ^ΑΑΧΑΧΑΑ.\_ΊΛχΑΧ.ΑΑ.%4ίΑ.Α

.. aVAAVQAAXAAAAAA'

V

WA AAA AX AAAAA \ΑΧΑΑΑΑν.·Ι>Λ<ί«·.'ι:ύΑΧΑ^ΧΑ\\ΑΑ\Α\\\ΑΧΑΧ\4ΛΖ>ΛΛ>ΧΑ.ΧΑ'άΐ^Α\ΑΧ\Α\'”&Λν.ν.>*χ5^ΑΧΑΧ'^1.\Α^Α'ΧΑΑΑΑ\Α\χ·&Χ

..... ' '· ' ........^b | ?

•4

AA^AA· '«A •J'^SS^A^^AAX^.^^KAi.XA^jS^A.s^xxA^^XAy·,;,^;^^ >Α\ΑΚ^ΑΑ1Α1Α>3ψ,νΖ' ^..· . - . - αΑΑ α^^,ΧΑΑΑΑΑΑΑΑΑΧ. J S's. -¼.. - . » ·ΑΑΑΑ*^α,ΑΑΑΑ· ^^*Λ*Χ·* ’ ^\AXAAA.AXAA^AAAAAW\4£z · ✓-·<-*·Α .^χΑΑίΑΑΧΑΑΑ^^ΑΑΑ'.*.. Α'Χ

Α^ΑΑΑχλχΑΧΑΧΑΑ A Ρ ΧΑΧΑΑΧΧΧΧΧΑΧΑΧΑΧΑΑ Ζ-* X ΑΑΑΧΑ^^ΑΧΑΧχΑ Ζ ^Α Α^ΑΑΧ'Χ^ΑΑΑ A ' ΑΑχΑΑ^ΑΑΑ%ΑζχΑΑΑ ✓ a Α Ά ιΑΧ^^αΑ.ΑΑΑΑάΆ^ «**^^*λίΑ A * A I »Α Al A^aAaAAAA '» '<^1 A A I aWA'A’A^AA'a AAAAAA^ λ^Α A A A lA lA S^B a A A A X\l A A AA aA A ::::.: : x: : : : : f .; : : > : : : : : x : :}::::::

^/$'^>\^Ν^^^ΥΑ^Α'Λ^Λ/^Λ^?^χΑχγ\^^\Α\^ΑΖν^ΧΖ/^»χ.ξ^ΑΑχΑ<\χΑ^ΑίΝΑ\'Α<«Λ>ή5ίί>ΖΖΖ,^.

^χ ΑΑΑΑΑΑ^χΑ A AAAa4aAAAXAXa ΑΑΑΑ^ζΒ AB * _z ' A IaIa 'A'A «•AA'aAAA-SiAASAAA^A^JbAAAA-i.-.«i'bX>x4^,'^AS'$b\\4j.A^.SAA\AA’.-.-.X».»'a'.|A.%%-b.^bAAAS A AAA XA^^^AXAAJ, A '· -----%¾.........^ΧΧΧΧΧΧΧ^,,.,. ........-4¾ . aXAXzAa, .. . ^χχχχ.-.-^f /Α>^^\Χ'^Αχ·Α^γ.ΑΛ^ν.Ζ;ί^Λ'.|’.Α'^\^ΑΑ\Χ'^Αχ4<^_;ΑΑΑ·.^.Α>.^Ζ'^<Α:._<ν<^Β,\χΑ?^ΑΑΑ^ΑΝΑ·.ν.^. AA V Α^ΑΑ ΑΑΑΑΑΑα'^ΧΧΧΑΑ f ** ^ΑΑΑΑΑΑΑΑΑ^, .ΑΑ-.Λ :-^^>>>Ν\\\\Α§^\χψχν^/-Λ>ΖΖ^>>^^\Α^\Α^χ\χ\χν -^ΛΆΌΧΧχί^χΑΑΑΑΑΑΑΆ;·

ΆΑΑΑΑΑΑΑ^Α-.Α^' ^Α\ΑχΑ·§ΧΒ^.ΧΧΧ^ .

... ',Ί,Ά”^Α|ΑΑΑΑΧΑΧ^%^ΒΜΖΖΑ*.*.*.* Λ

Α^

~.α|._Λ>><ΑΑΑΑ^.ΑΧ}§^ λ*λζΛ<_ζ . .· ι'.Α.α4Α.....A

Τ _ύ <ΑΧ .

FIG. 2

WO 2017/037384

PCT/FR2016/052157

2/13

C

D

C

D

FI/' Q

WO 2017/037384

PCT/FR2016/052157

FIG. 5

WO 2017/037384

PCT/FR2016/052157

4/13

FIG. 6c

WO 2017/037384

PCT/FR2016/052157

5/13

FIG. 7a FIG. 7b

FIG. 7c

WO 2017/037384

PCT/FR2016/052157

6/13

WO 2017/037384

PCT/FR2016/052157

7/13

M6	CHA .............................. ..............Ϊ...............		L6			K6
	i	©			©	ί ί ί
	! ©			©		i ί t ϊ
M7	1 1 1 1 1 1 1 1 1	©	L7		©	(Κ7 ϊ ί ί ϊ ί ί
M8	! ©		L8	©		ί ι t Κ8

FIG. 11a

FIG. 11b

WO 2017/037384

PCT/FR2016/052157

8/13

CHA

WO 2017/037384

PCT/FR2016/052157

9/13

Ζοπ1

FIG. 14

FIG. 15

WO 2017/037384

PCT/FR2016/052157

10/13

ST1

ST12

ST13

FIG. 16

WO 2017/037384

PCT/FR2016/052157

11/13

	·< .........
m:		<&;/ /,-.,-,¾
	J · <”O	4^· ' ί- .··'
w		ν·'·ν.ν.·.»
		·* __ 77777 /:/:://:/:: :/1/:/0:087/::
4» j-l*’ *	J*; . iU' <.,?'	ν^ίΚί. *1 ' L-. '. ¢,
	rj?AW5?3Si	.. ... \-x
w	O,'.'	* · 'W>·,·
......JN........
<*	’'-•4 φ 7 / 7 //^,/5^^//// / 7 7 77/ 7 7 /
	xx^· ·. ’ <: < ............................................	«h-'ΐ -
/7//^
l	*77,,	1!

FIG. 17

	/777/077707777//777/:7/777::: :/7
~		«c /,7...,. .. W·! 7 7 ·'
-	ά^··.· <· 4 4' ·'< *· *· ·“ ί s»M ^Wl. ’.»s ·.<''.'
	% ..V.'X Λ, 'W ’ ,.	dfe 7 - - .
		77// W’ : -z:.,-,sx. “* ··x v\\ amt·_ ί·^·ί' <
A	xa*£ x .»; ,«·· \ <-ίΛχ^^^χ/ΐΛ.Α
		W' . , X lib <?/..> >;
	M SSOii^i· Wfr\ <»vx> >:
	777 ? ^W7: ^^7-^7^:^-¾. 7: /	3$SK ί·», :>’.,· ^.,...-. .;. λ ,<-... .....Wx ί\\:> :·,
*	7 4' 4' 4 iQ 1» >	ζκ ca^-'> /:/7/7/ .:
-
	1*·-··'· 1_1	.t<#'-7-

FIG. 18

WO 2017/037384

PCT/FR2016/052157

12/13

FIG. 20 wo ²°17/037384

FIG. 21

WO 2017/037384

PCT/FR2016/052157

METHOD FOR MAPPING FUEL ASSEMBLIES

The present invention relates to a determination of positions of objects such as fuel assemblies of a nuclear reactor vessel. It relates in particular, but not exclusively, to safety applications for the fuel assemblies of a nuclear reactor.

In a nuclear reactor vessel of a power station, fuel assemblies forming a fuel core are periodically inspected and replaced. To this end, fuel assemblies must be unloaded from the fuel core and new ones must be loaded. Once the new fuel assemblies are loaded, their correct positioning must be verified before closing the head of the vessel, the head comprising Upper Internals (UI). The UI include pins which protrude and are intended to be inserted into corresponding housings of the fuel assemblies, called S-holes. The S-holes are the upper nozzle elements of the assemblies, into which fit the centering pins of the top plate of the core (see Figure 1).

A set of nominal positions is ideally defined for the fuel assemblies. However, it is possible for a fuel assembly to be offset relative to its nominal position beyond a certain capacity for recentering, which leads to the possibility of the assembly catching during placement of the UI. Such an offset can cause forceful entry of the UI pins into the S-holes of the fuel assemblies. Although this does not interfere with the operation of the nuclear reactor, the fuel assemblies may remain caught on the UI the next time the vessel is opened, resulting in long and expensive maneuvers to free them and a potential safety hazard for the facility. In addition, the fuel assemblies are submerged in water which makes the activities of securing and unseating the assemblies even more difficult.

A method has been proposed in document FR-2999011 for accurately determining a position for each S-hole of the fuel assemblies before placement of the UI, to enable intervention in the event of too large of an offset of one of the assemblies relative to a nominal position (and therefore the possibility for it to become stuck in a UI). Due to various constraints, particularly to the fact that the fuel assemblies are submerged in water, it remains difficult to accurately estimate the position of the S-holes. Turbulence due to inhomogeneities in the liquid medium (local temperature differences) further increase the complexity of such estimates.

In the following, we are therefore focusing on estimating the position of the S-holes in particular, of the fuel assemblies after reloading. More particularly, official nuclear safety procedures require estimating the positions of S-holes by manual measurement of the relative gaps between assemblies.

These gap measurements are carried out by video inspections conducted by one or more operators

WO 2017/037384

PCT/FR2016/052157 (people). These measurements take a long time and the resulting uncertainty depends on the person carrying out these measurements, particularly their experience.

The invention aims to improve this situation.

For this purpose, it proposes keeping the intervention of a user (a person) in order to satisfy the aforementioned official procedures, this user viewing the S-holes in one or more images captured by a camera and inspecting whether their position matches the expected position. However, the invention improves the reliability of the information provided to the user for performing this verification. For this purpose, it provides computer processing which applies shape recognition to the captured images in order to locate the S-holes and also to display visual cues on the captured images which enable a user to identify the S-holes with certainty and thus to inspect their position.

However, as previously stated, the images of the fuel assemblies are very noisy, mainly because of turbulence related to temperature differences in the water. In one approach of the invention, rather than capturing a single overall image of the assemblies in a wide-angle view, the images are captured by moving the camera closer to the assemblies and having the camera follow a predetermined path along the assemblies in a top view. For example, rather than position the camera for a wide-angle view and then performing an optical or digital zoom towards the assemblies, the camera is moved physically closer to the assemblies to obtain close-ups (where of course it is then possible to adjust the apparent size of the assembly elements when necessary, using the camera's zoom).

However, denoising all the images successively captured along said predetermined path takes a long time and the processing times are unacceptable for an application in semi-automatic real time.

The invention more particularly proposes limiting the denoising to the areas of interest in the captured images (hereinafter also called marked regions), these areas of interest containing the expected position of the S-holes with a given tolerance.

The invention therefore relates to a method for verifying the positions of fuel assemblies arranged within a nuclear reactor vessel, each fuel assembly comprising at least one reference element, the method comprising the steps of:

- capturing images from a camera filming at least the reference elements of at least a portion of the assemblies,

WO 2017/037384

PCT/FR2016/052157

- denoising the captured images,

- image processing to locate the reference elements in the denoised images.

In particular, the camera fdms in close-up of only a portion of the assemblies in each image, and moves along a predetermined path in order to obtain a sequence comprising consecutive images of all the assemblies.

The steps of image denoising and processing are then carried out in at least one area of interest defined in each captured image. This area of interest includes one or more expected positions of reference elements, with a tolerance around said expected positions so as to also include the actual positions of the reference elements in the captured images. The area of interest is thus determined in each image based on said predetermined path.

As previously stated, such an arrangement saves processing time for the denoising, but also for the processing which locates reference elements in the areas of interest in the captured images. The reference elements may be S-holes, as indicated above, but possibly also other elements such as the baffle comers, as described below.

In one particular embodiment, the method further comprises the generation of at least one virtual marked region defining the area of interest and superimposed on a control screen showing the actual images captured by the camera, for verification by an operator. In this embodiment, the method further comprises the step of:

- after determining the location of at least one reference element in the area of interest, displaying an identifier of the located reference element together with said marked region, the identifier being predetermined and selected from a table of previously saved identifiers.

The processing method according to this embodiment thus follows a list of reference element identifiers corresponding to the reference elements expected along the predetermined path of the camera.

It will thus be understood that such an embodiment is advantageously suited for a semi-automatic procedure. The operator can check against a provided document that the displayed identifier is indeed

WO 2017/037384

PCT/FR2016/052157 the one indicated in the document and lies within a series of successive identifiers along the camera path.

In one particular embodiment, the step of denoising the captured images comprises a reduction of at least the optical effects of turbulence related to local temperature variations in a fluid in which the assemblies are immersed. One possible embodiment of such processing, which has provided satisfactory results, is disclosed in EP-2668635.

In one embodiment, the image processing step implements shape recognition to locate the reference elements. Particularly for reference elements comprising holes (S-holes) intended to accept securing pins between assemblies, the shape recognition may advantageously comprise a Hough transformation to locate the circles delimiting the holes. As discussed below, the recognition of circular shapes based on Hough transformation is particularly robust and suitable, including in images which are initially very noisy and for which the denoising still leaves some noise (even if greatly reduced).

The predetermined path may also include one or more assembly baffle comers, such that the areas of interest in the captured images include the baffle comers. The method may thus further comprise image processing to determine the locations of baffle comers in the denoised images, in addition to reference elements such as the S-holes described above. For example, the locations of the baffle comers can be determined by shape recognition (typically a contrasting rectangular shape in the images, as discussed in the description below). However, this location determination is preferably confirmed by an operator via a user interface (computer mouse, keyboard, or other).

The positions of the baffle comers are in fact used to place the reference elements within an absolute reference system of a map of the assemblies. Thus, the detected baffle comers allow determining the origin of this reference system and the positions of the S-holes are given by their absolute coordinates (in the x, y plane) within this reference system.

The method may further comprise generating a warning if at least one of these positions deviates from its nominal position by more than a predetermined threshold.

The camera preferably moves in translation in a plane parallel to a central plane of the assemblies, the predetermined path of the camera then following a winding path along the successive rows of

WO 2017/037384

PCT/FR2016/052157 assemblies. Here, the term rows is of course understood to mean horizontal rows as well as vertical rows, relative to the orientation of the camera.

The method may further comprise a step of mapping the assemblies wherein the actual relative positions of the assemblies, obtained from the determined locations of the respective reference elements in the captured images, are determined relative to their respective expected positions. Of course, this mapping is advantageous for determining all the discrepancies relative to the expected Shole positions. However, it is optional and is not essential to the semi-automated procedure for monitoring video images by an operator.

Advantageously, the method is assisted by a computer module for monitoring the images captured by the camera along its path, in order to locate the area of interest in the captured images with respect to an initial starting position of the camera, and based on its predetermined path.

This computer module executes a computer program to carry out these steps, and the present invention also relates to such a computer program, comprising instructions for implementing the above method when the program is executed by a processor.

An exemplary flowchart of such a computer program is illustrated in Figure 16, discussed below.

The present invention also provides a device for verifying the positions of fuel assemblies arranged within a nuclear reactor vessel, each fuel assembly comprising at least one reference element, the device comprising in particular:

- a memory MEM (Figure 21) for storing captured images from a camera fdming at least the reference elements of at least a portion of the assemblies,

- a computer system CI (Figure 21) cooperating with the memory, for denoising the captured images and determining the location of reference elements in the denoised images.

In particular, the camera fdms close-ups of only a portion of the assemblies in each image and moves along a predetermined path to obtain a sequence comprising consecutive images of all the assemblies. The computer system then determines an area of interest in each captured image based on said

WO 2017/037384

PCT/FR2016/052157 predetermined path, the area of interest including one or more expected positions of reference elements, with a tolerance around said expected positions so as to include the actual positions of the reference elements in the captured images. The computer system then performs the denoising and determines the location of reference elements only in said area of interest, and does this for each image.

Other features and advantages of the invention will be apparent from examining the following detailed description and the accompanying drawings in which:

- Figure 1 shows a fuel assembly and two S-holes (TRS) which serve as said reference elements,

- Figure 2 shows a winding path of the camera along a nuclear reactor core of 900 MWe,

- Figure 3 illustrates the operating principle of the location determination by marked regions, shown here in a row of index A (at the edge of the set of assemblies)

- Figure 4 illustrates the automatic detection and identification of S-holes,

- Figure 5 illustrates the denoising applied to an image containing turbulence (denoising with σ = 8)

- Figures 6a, 6b and 6c show the preprocessing of an image contained in a marked region a): before edge detection, b): with contrast enhancement, and c): with denoising,

- Figures 7a, 7b and 7c compare the respective effectiveness of different types of edge detection,

- Figure 8 illustrates the votes collected in a two-dimensional accumulator (a,b) in the context of Hough transformation,

- Figure 9 illustrates the votes collected in a three-dimensional accumulator (a,b,r), each edge point voting for a cone (or cone portion) of which the tip is taken, in the context of Hough transformation,

- Figure 10 illustrates the votes collected along two circular arcs on both sides of the respective directions of a gradient GRAD in order to identify the accumulation point ACC as the center of the recognized circle, in the context of Hough transformation,

- Figure 11A illustrates the typical content of an image with S-holes to be identified by Hough transformation, and Figure 11B is one example of a real corresponding field of view to be observed when mapping,

- Figure 12 schematically illustrates a captured image requested in order to create a map, here including baffle edges,

WO 2017/037384

PCT/FR2016/052157

- Figure 13A schematically illustrates a captured image requested in order to create a map, here including a lateral baffle edge, and Figure 13B is an example of a real corresponding image including baffle edges,

- Figure 14 illustrates a fine-tuning of circles detected by cross-correlation, for fine-tuning the determination of the position of an S-hole,

- Figure 15 illustrates a fine-tuning, here manual, of the positions of the baffle comers, in two different images presented here as examples,

- Figure 16 summarizes the steps of a method according to one possible embodiment of the invention,

- Figure 17 illustrates clusters of S-holes detected on a portion of the core, as an example,

- Figure 18 illustrates the clusters of Figure 17, after elimination of outlier circles (false positives for S-holes),

- Figure 19 is an exemplary map of the offsets of S-hole positions relative to the nominal positions,

- Figure 20 illustrates the estimated mapping results for some assemblies, with the properly positioned assemblies in white, those with a deviation from the nominal that exceeds 6.5 mm in gray, and those for which the deviation is greater than 7.5 mm in black,

Figure 21 illustrates a device for implementing the method according to an exemplary embodiment of the invention.

A fast and accurate semi-automatic method is proposed here which can yield up to a complete mapping of the fuel assemblies, based on image processing. The method may include three main steps:

Capturing video sequences: the camera is moved in translation in a plane parallel to the plane of the assemblies,

Processing the videos: First the images are restored, to eliminate the effects of turbulence among other things. Then image processing algorithms are used to locate and identify S-holes in a semiautomatic manner. Other reference elements are also located in the images (for example baffle comers identified by the intersection of their edges),

Mapping: optimization processing is then applied in order to obtain a complete map of the assemblies of a given reactor core. This processing makes it possible to calculate the positions of each S-hole automatically, while taking into account the geometric constraints such as distances between centers (distance between two S-holes of the same assembly), the non-overlapping of assemblies, and distances between baffles.

WO 2017/037384

PCT/FR2016/052157

The method of the invention was tested at a site where the positions of the S-holes were precisely known. In the different passes over the rows and columns, application of the method gave a maximum error of 1.7 mm and a mean error of 0.6 mm, with 95% of the errors being less than 1.2 mm. The total duration, including the image acquisition time and image processing time, was less than three and a half hours with only one operator for the video processing, and less than two hours with two operators.

The method is now detailed below.

Referring to Figure 1, we wish to obtain a series of images containing S-holes (denoted TRS). For this purpose, a camera moves in translation parallel to the plane of the assemblies having these holes in their uppermost portion, along a winding path following the rows or columns (Figure 2). The camera must be centered on the current row (or column) with a margin of about 50 pixels to the left and to the right in order to observe the S-holes of neighboring assemblies. An example of a suitable field is shown in Figure 4.

It is not necessary' to wait until all the video sequences have been captured before beginning the image processing steps described in detail below. Once the first sequence has been captured (the first row of assemblies in Figure 2), the video can be sent to a processing device in order to begin processing while the camera travels along the second sequence, and so on.

The S-hole search areas in the images are defined by a set of four areas of interest called marked regions which are superimposed (virtual images) on the actual images captured by the camera, and then follow the movements of the camera by moving along with the translation of the camera (apparent displacement in the captured video images). These marked regions allow:

• Introducing initial information on the objects to be detected in each marked region (amount and identification number of S-holes to search for, presence or absence of baffle comers), • reducing false detections by means of restricted search areas focused on the S-holes: as shown in Figure 4, there are many circles to avoid at the center of the assemblies, • and also reducing the computation time by not processing the entire image. The four marked regions can be processed in parallel on a computer having at least four processors.

The principle of the marked region analysis operation is illustrated in Figure 3.

The four marked regions (denoted A, B, C, and D) are initialized by two clicks of the user on the comers of the first assembly at the beginning of each sequence. They are fixed relative to each other

WO 2017/037384

PCT/FR2016/052157 (and move as a block along the vertical axis) as they follow an estimation of the translational movement of the camera.

In the example below, a path is followed along row A (A9, A8, A7, ...), shown in Figure 3. At the beginning of the sequence, there are initially the four marked regions A, B, C, D. Then, as the camera moves downward (dotted arrow), the marked regions appear to move upward (gradually leaving the camera's field of view CHA represented here with dotted lines, by way of illustrative example). Marked regions A and C disappear from the image at the top. They reappear at the bottom (respectively at the left and right) of the image when the camera reaches the center of assembly A8. At each reappearance of a basic marked region (A, B, C and D), a marked region counter c is incremented, which is used to update the objects of the search. For example, marked region B contains reference element A9S1 (hole S1 of assembly A9) when the counter c_B = 1. It then contains element A7S1 (hole SI of assembly A7) and baffle comer A7 when the counter c_B = Ί. The successive contents of the basic marked regions A, B, C, D together form the configuration of a sequence.

Because the estimation of the camera translation based on the recognition of reference elements is subject to drift, the position of the marked regions is realigned whenever two circles are simultaneously detected with an expected angle between them (within a margin). If desynchronization of the marked regions relative to the S-holes occurs, the user can reset them manually using a mouse or other computer input device (keyboard, touchscreen, or other).

The following provides details on an example implementation, referring again to Figure 2. The diagram in Figure 2 corresponds to the exemplary case of a 900 MWe reactor core but could easily be transposed to the case of a 1300 MWe or N4 reactor. The described path is by row, but it can be replaced by a path by column if so required by the implementation.

The path over the core snakes along each row in turn (Figure 2): o Sequence A = row A from left to right o Sequence B = row B from right to left o Sequence C = row C from left to right o Sequence R = row R from left to right

The camera moves in translation parallel to the plane of the assemblies. It must not make any other movement (rotation, altitude variation, etc.). The camera moves relatively slowly (a few centimeters per second). It is not necessary to stop the camera as it travels along a row. However, there should be a

WO 2017/037384

PCT/FR2016/052157 pause of a few seconds at the ends of each row (beginning and end). Each row is recorded in a different sequence, without image compression. Thus, in the example shown, fifteen video files respectively correspond to the recordings of fifteen passes by row or column.

Preliminary adjustment of the camera is performed one time only, before any of the sequences are recorded. It is done by placing the camera vertically above the center of an assembly not adjacent to baffles and containing a control rod. The purpose of this preliminary adjustment is to fix the orientation, altitude, and zoom factor. For the orientation, the axis of the camera must be vertical. The camera is placed above the axis of the control rod which must appear vertical in the image. The camera must be such that:

- the short side of the image (576 pixels for PAL format) is parallel to the direction of travel of the camera (therefore parallel to the rows of the core),

- the long side of the image (720 pixels for PAL format) is perpendicular to the direction of travel of the camera (therefore parallel to the columns of the core),

- the row traveled is vertical in the viewed image.

Regarding the altitude, the camera must be as close as possible to the assemblies (typically 0.8m is the average distance in the shots presented here as examples).

At the fixed altitude, the zoom factor should allow obtaining the configuration given in Figure 11 A: when the camera is located directly above the center of an assembly, its field of view should include the entire assembly at the center, as well as hole SI of its left neighbor plus a margin (about 50 pixels) and hole S2 of its right neighbor plus a margin (about 50 pixels) as shown in Figure 1 IB. This level of zoom ensures the simultaneous presence of four S-holes at the bottom or top when the camera travels over rows B to P. Referring to Figure 11A, when the camera is directly above the center of an assembly (assembly L7 here), the image should contain the entire assembly L7 at the center, and should provide a simultaneous view of hole S1 of assembly M7 and hole S2 of assembly K7 (with a margin of 50 pixels on either side).

At the start and finish of a row, the baffles should be fully visible with a margin MAR that is as large as possible from the edge of the image (Figure 12). Deflection of the camera should be avoided. It preferably remains parallel to the plane of the assemblies (the optical axis of the camera thus remaining preferably perpendicular to the plane of the assemblies), keeping it motionless for at least five seconds at the beginning and end of each sequence, as illustrated in Figure 12. Thus, referring to Figure 12, when the camera is at the edge of a row (beginning or end of sequence), here the end of row H, the viewed image must completely contain the two baffle edges BAFF, with the largest possible margin MAR.

WO 2017/037384

PCT/FR2016/052157

In row A (respectively R), located at the edge of the vessel (example of Figure 13 A), the fdmed scene must include the row of assemblies traveled but also the neighboring S-holes (with a margin of 50 pixels) located in the space to the left (respectively to the right) as well as the baffles BAFF located to the right (respectively to the left).

Thus, referring to Figure 13A, when the camera is over the edge rows (here row A), the captured image must completely contain the two baffle edges and the three S-holes of assemblies A8 and B8: here SI and S2 of A8, as well as SI of B8 with a margin of 50 pixels. The camera must remain in translational movement parallel to the plane of the assemblies. No deflection should occur.

Figure 13B shows an example image captured in this manner at the vessel edge.

The lighting should be as uniform as possible.

Each marked region is subjected to image denoising in order to improve circle detection when there is thermal turbulence (as illustrated in Figure 5 showing strong turbulence outside the marked regions). The processing power (parameter σ of EP-2668635) can be adjusted by the operator for easier monitoring. The processing power σ should be as low as possible while the holes remain visible, as a high σ enlarges the outlines, which introduces greater uncertainty in determining the location of the circles.

If the circle detection in the marked regions is automatic, their identification occurs semi-automatically due to the two vertical lines Li and L2 visible in Figure 4, which the user can move with a mouse or keyboard keys (left and right arrows). They allow classifying the detected S-holes as Si (denoted TRS1) or S2 (denoted TRS2).

The next step, edge detection, is crucial to the quality of the mapping.

The quality of the edge map heavily impacts the determination of circle locations in the Hough transform-based detection step, and thus impacts the overall accuracy of the processing. A good edge map must be repeatable: the edges detected in the same marked region over time should have little variability (oscillations of the detected circles may be observed otherwise).

The image is first converted to grayscale (Figure 6a), then its contrast is increased (Figure 6b). To avoid false edge detection related to the presence of noise, a bilateral filter is then applied (Figure 6c) which reduces noise while maintaining sharp edges. This preprocessing is illustrated in Figure 6.

For the edge detection, we can cite such processing as is described in:

WO 2017/037384

PCT/FR2016/052157

C. Tomasi and R. Manduchi, Bilateral filtering for gray and color images, IEEE Sixth International Conference on Computer Vision, pp. 839-846, 1998,

P. Meer and B. Georgescu, Edge Detection with Embedded Confidence, IEEE Transactions on Pattern Analysis and Machine Intelligence, 23(12), pp. 1351-1365, 2001, which have provided satisfactory results as illustrated in Figures 7b and 7c in comparison to Figure 7a showing poorer results in the circumference of the S-holes and more introduced artifacts. Edge detectors according to the above references give fairly similar results, with connected and smoother edges.

The computation times for the edge maps of Figure 7 are 3 ms for the processing in 7a, 96 ms for the processing in 7b (Tomasi et al), and 10 ms for the processing in 7c (Meer et al) for the same hardware configuration. The computation time is a decisive factor, because the verification of assembly positions must be done quickly during the operations of loading and unloading fuel assemblies in the core of the reactor, due to safety issues. The preferred processing is therefore 7c.

For circle detection in a binary image, the circle Hough transform is used. This processing consists of votes from each point of the edge map for all hypothesized circles that pass through this point. The point that has the most votes is considered the center of the most probable circle. To detect circles of radius R of equation: (x — a)² + (y — h)² = R², the votes are collected in a two-dimensional accumulator (a, b), as illustrated in Figure 8 (in the area defined by dotted lines) .

For a given radius R, each edge point votes for the possible centers of the circles to which it may belong: it votes for a circle of center R around it. The accumulation points correspond to hypothesized circles in the image. However, we do not know the exact radius R but have an interval [Ri; R₂] (which is determined by adding a margin to the radius measured in the preliminary adjustments). The accumulator then has three dimensions (a, b, r) and each edge point votes for a cone (or a part) of which the tip is taken (Figure 9). The accumulation points (intersection point of cones) provide the hypothesized circles (center and radius).

Thus, referring to Figure 1, if the radius is unknown, each edge point votes for a cone in the Hough space.

In fact, it is not really necessary for each edge point to vote for an entire circle. The gradient at point A of the circumference of a perfect circle of center O is oriented along a vector OA. It is therefore sufficient to vote on each side of the direction of the gradient (with an angular deviation to incorporate

WO 2017/037384

PCT/FR2016/052157 the uncertainty), which makes two circular arcs of votes (as shown in Figure 10). In Figure 10, each edge point votes for two circular arcs, one on each side of each gradient direction GRAD.

This vote restriction is significant in reducing the number of unnecessary votes which can cause false positives.

The local maxima of the accumulator of the Hough transform correspond to hypothesized circles (center and radius). These hypotheses remain to be validated to verify that they actually correspond to circles of the image. For this, use is made of the technique called a contrario validation, which proposes statistical tests for detection that are based on principles of human perception:

A. Desolneux, L. Moisan and J.-M. Morel, From Gestalt Theory to Image Analysis: A Probabilistic Approach, Springer-Verlag, Interdisciplinary Applied Mathematics, vol.34, 2008.

The maximum from the accumulator of the Hough transform is not very accurate for locating the center of the circle, and there may be several high values nearby (false positives). These circles must be located with the greatest possible precision, however (to the nearest pixel). For the cameras used (PAL format 720x576) and the recommended field of view, we typically have a resolution of about 0.5 mm per pixel. To obtain positions to the nearest pixel, the detected and validated circles are fine-tuned by cross-correlation with a perfect circle model. For this purpose, an area ZON1 around the detected circle is extracted (Figure 14) and the edge map is cleaned to leave only a ring COUR1, as illustrated in Figure 14. We then search this area COUR1 for the maximum correlation with a perfect circle model MCER of a radius corresponding to that found by the Hough transform, in order to refine the determination of the S-hole TRS.

The reference points for positioning the fuel assemblies in an absolute reference system are the inner comers of baffles (intersection of baffle edges). When processing the video images, an approximate position is automatically recorded for each baffle comer. In fact, the baffle comers always appear inside a marked region and it is known which marked region is to be searched for which baffle comer. To be incorporated into the mapping process, the baffle comers must be detected simultaneously with at least one circle in the image. For the baffle comer detection, the approximate position of the baffle comer is therefore chosen as the center of the marked region where it is visible with the maximum number of circles in the same image. For a precise readjustment in absolute position, in the final map, these positions must be fine-tuned. After the video sequences have been processed, the user is

WO 2017/037384

PCT/FR2016/052157 presented with zoomed images centered on the approximate positions of each baffle comer. The user can then click on the precise position of the baffle comer as shown in Figure 15. Several difficulties may arise, however:

- The baffles are not quite perpendicular, which can introduce ambiguity in the position of the comer,

- The interlacing of the videos may be corrected (for example by a temporal filter), but this introduces a loss of resolution,

- Low contrast related to turbulence.

Referring to Figure 16 illustrating the main steps of the above method, after a first image acquisition step ST1, the images are denoised in the areas corresponding to said marked regions A, B, C, D in step ST2. Then, a Hough transformation is applied in step ST3 in order to identify the positions of the Sholes in step ST4. In the denoised areas, shape recognition can be implemented in step ST5 in order to identify the baffle comers. In addition, an operator can intervene in step ST6 in order to determine the baffle comers more precisely in step ST7 because they can then provide an absolute position in the mapping of S-holes, as seen below. Next, the processing device can prepare a display of reference elements in step ST8 (S-holes and baffle comers) to help guide the operator in the images presented to the operator on a display screen TVS (Figure 21). To this end, the processing device may use in step ST9 a list of successive reference elements successively filmed by the camera, based on its path predetermined in step ST10. Thus, in step ST 11, an overlay is displayed for the operator who can then view it in step ST12. Lastly, the map CARTOG of all the assemblies is constmcted in step ST13.

The above procedure for the automatic detection of S-holes may yield false positives, even after a contrario validation. Outlier circles must therefore be detected and eliminated before constmcting the map. For this purpose, clusters of all S-hole detections are formed for each sequence. The position of an S-hole detected in image n is written:

(Xts ^— ^camera I/¹] <Z X_Objet [n] ( Frs ⁼ ^camera W + <Z Yobj_et [ⁿ] where (XcameraM, KameraM) is the estimated position (in pixels) of the camera obtained by summing the estimated displacements up to image «, (X_Objet[ⁿL YobjetM) is the position of the detected S-hole (in pixels) in image n, and a is the correction factor for the parallax effect:

^camera ^TS ^a ~ Z '' ' camera

The parallax factor can be estimated by a least squares solution.

WO 2017/037384

PCT/FR2016/052157

Once the clusters of points are formed for each sequence (for example as in Figure 17), the circles (x,

y) that satisfy the following two conditions are retained for each S-hole:

(t) |x — mediane_x\ < Seuil_x (ii) |y — mediane_Y\ < Seuil_y where Seuil means threshold, typically with Seuil_x = Seuil_y = 15 pixels for example. Conditions (i) and (ii) exclude circles that deviate more than 15 pixels from the median in X and Y among the circles detected for a same S-hole. An absolute threshold of 15 pixels is chosen (corresponding to about 8 mm) rather than a statistical limit (such as 3 a_exp) because the estimate of the experimental variance can be poor if there are few detections. For example, hole L12S2 of Figure 17 is detected six times and one can clearly see an outlier point which is far apart from the cluster of other points. This outlier substantially increases the experimental variance in X, and thus the criterion:

|% — mediane_x\ < 3 a_expX does not allow rejecting it even though it is more than 15 pixels from the median.

The corrected clusters of the example are shown in Figure 18.

It is important to mention here that the estimation of the camera path is not used to calculate the positions of the S-holes, because it is subject to uncontrolled drift over time. The path estimation is only used for the adjustment of marked regions and the elimination of false positives. It is also sufficiently precise in the short term for the creation of point clusters for the same S-hole.

The mapping technique proposed here is based on estimating distances between S-holes detected in the same image. For each pair of S-holes considered, the distances between S-holes (in mm) are expressed according to the coordinates of the S-holes detected in the image (in pixels) while taking into account the differences in altitude between assemblies and a possible deflection the camera.

The problem is determined by multiple inputs, particularly because the same pair of S-holes is seen in multiple images. To take into account all available information, the problem is solved by minimizing a least squares criterion. The optimization criterion also takes into account all geometric constraints such as constraints on spacing between axes, the non-overlap of assemblies, the distance between baffle edges.

Solving the optimization problem yields a relative mapping, meaning that the positions of the S-holes are estimated to the nearest translation and orientation. Two possibilities are then considered for transitioning to an absolute mapping.

WO 2017/037384

PCT/FR2016/052157

The first approach is very simple and involves post-processing the results of the relative mapping. The idea is to find the affine transform which minimizes the sum of the mean squared deviations between the estimated positions of the S-holes and the nominal positions of the S-holes. This is called calibrating to the nominal. The underlying assumption is that the mean of the S-hole deviations at the nominal positions is zero. This seems realistic when the mean of the deviations between the estimated positions and the nominal positions is less than 0.5 mm, which has been observed in some tests. However, this assumption is less valid if a large number of assemblies are greatly offset, resulting in a significant deviation (1 mm) between the mean of the actual positions and the mean of the nominal positions.

The second approach is to make use of elements whose position is known in the absolute reference system, for example the comers of the baffle edges. The idea is to determine the location of these elements in the video sequences (as described above) and to include additional information in the optimization procedure concerning the distances observed between S-holes and comers of baffle edges. In the expression of the optimization criterion, the positions of the comers of the baffle edges take the values of the nominal positions. One advantage of this approach is that it allows directly estimating absolute positions.

Of course, uncertainties concerning the actual position of the baffle edges (known with an uncertainty of about 0.5 mm) or their location in the sequences (same order of magnitude as the error in the location of the S-hole circles) may appear. One then needs to determine the location of as many comers of baffle edges as possible in the video sequences in order to minimize the impact of these uncertainties.

It should be noted in any case that the method according to either of these embodiments is more reliable than a method which only provides a single reference point (in a single overall image) for locating the S-holes in an absolute reference system. Such a method has a major disadvantage: the positions of all the S-holes are dependent on the determined location of this single point. A mistake in determining this location (a click error by the operator or an automatic location determination error) affects all S-holes.

Instead, a plurality of reference elements are chosen here, which are the points located at the intersections of the partitions (for example 44 baffle comers in a 900 MWe core). This solution has the advantage that the absolute positioning of the S-holes is not dependent on a single reference element.

The final optimization is not determining the location of all the holes relative to one reference point but to a set of reference points, which is much more robust.

WO 2017/037384

PCT/FR2016/052157

The use of a camera filming video images can be justified by:

- the computation time: the camera advances in a translational movement over the assemblies, preferably without stopping,

- the number of images detected per assembly: the camera captures 24 frames per second, which allows (in contrast to a single captured image of a set of assemblies) using a plurality of images per assembly for the detection of S-holes.

Next, the image processing step which eliminates thermal turbulence advantageously reduces the uncertainty in positioning the S-holes that was introduced by the turbulence.

Once all the locations of the S-holes in the images (pixel coordinates) have been determined, the entire core is reconstructed by a constrained optimization process that can provide a geometrically consistent solution. The final position of all the S-holes is such that:

no assembly overlaps its neighbor,

- no assembly extends beyond the tank,

- the distance between two S-holes of a same assembly (center to center) is respected.

The absolute positioning is relative to the detected baffle comers.

Geometric constraints are taken into account, preferably during the optimization which results in the mapping presented below, and the processing performed in the optimization may be as described in

FR-2999011.

The obtained mapping results are represented in graphical form, with the nominal deviation vectors, in Figure 19.

One can also represent the map of assemblies as shown in Figure 20:

- properly positioned (white, for example with a deviation of less than 6.5 mm), near the limit (gray, for example with a deviation of between 6.5 mm and 7.5 mm), and non conforming (black, for example with a deviation of more than 7.5 mm).

Of course, it is possible to generate a warning if at least one deviation is greater than a predetermined threshold (for example 7.5 mm in the example of Figure 20).

WO 2017/037384

PCT/FR2016/052157

A device for implementing the invention is represented in Figure 21. The device comprises a memory MEM for storing images captured by the camera CAM. The camera films the assemblies in a top view in order to obtain the S-holes (TRS) and the baffle comers CdB within the images. The device has a processing circuit CI comprising a processor PROC which retrieves the captured image data, and a working memory MEM2 storing instructions of a computer program according to the invention, in order to execute the above method. The processor PROC also cooperates with a graphical interface INTG connected to a display screen TVS, as well as with a user interface INTH connected to an input means such as a keyboard and/or a computer mouse SOU (a touch screen or other).

WO 2017/037384

PCT/FR2016/052157

Claims (13)

Claims

1. Method for verifying the positions of fuel assemblies arranged within a nuclear reactor vessel, each fuel assembly comprising at least one reference element, the method comprising the steps of:

- capturing images from a camera filming at least the reference elements of at least a portion of the assemblies,

- denoising the captured images,

- image processing to locate the reference elements in the denoised images, characterized in that the camera films close-ups of only a portion of the assemblies in each image, and moves along a predetermined path in order to obtain a sequence comprising consecutive images of all the assemblies, and in that the steps of image denoising and processing are carried out in at least one area of interest defined in each captured image, the area of interest including one or more expected positions of reference elements, with a tolerance around said expected positions so as to also include the actual positions of the reference elements in the captured images, said area of interest being determined in each image based on said predetermined path.

2. Method according to claim 1, wherein the method further comprises the generation of at least one virtual marked region defining the area of interest and superimposed on a control screen showing the actual images captured by the camera, for verification by an operator, the method further comprising the step of:

- after determining the location of at least one reference element in the area of interest, displaying an identifier of the located reference element together with said marked region, the identifier being predetermined and selected from a table of previously saved identifiers.

3. Method according to one of claims 1 and 2, wherein the step of denoising the captured images comprises a reduction of at least the optical effects of turbulence related to local temperature variations in a fluid in which the assemblies are immersed.

WO 2017/037384

PCT/FR2016/052157

4. Method according to one of the preceding claims, wherein the image processing step implements shape recognition to locate the reference elements.

5. Method according to claim 4, wherein the reference elements comprise holes to accommodate securing pins between assemblies, and wherein the shape recognition comprises a Hough transformation to locate the circles delimiting said holes.

6. Method according to one of the preceding claims, wherein the camera moves in translation in a plane parallel to a central plane of the assemblies, the predetermined path of the camera following a winding path along the successive rows of assemblies.

7. Method according to one of the preceding claims, wherein, as the predetermined path includes one or more assembly baffle comers, the areas of interest in the captured images include the baffle comers, and wherein the method further comprises image processing to determine the locations of baffle comers in the denoised images.

8. Method according to claim 7, wherein the locations of the baffle comers are determined by shape recognition and confirmed by an operator via a user interface.

9. Method according to one of claims 7 and 8, wherein the positions of the baffle comers are used to place the reference elements within an absolute reference system of a map of the assemblies.

10. Method according to one of the preceding claims, assisted by a computer module for monitoring the images captured by the camera along its path, in order to locate the area of interest in the captured images with respect to an initial starting position of the camera, and based on its predetermined path.

11. Method according to one of the preceding claims, further comprising a step of mapping the assemblies wherein the actual relative positions of the assemblies, obtained from the determined locations of the respective reference elements in the captured images, are determined relative to their

WO 2017/037384

PCT/FR2016/052157 respective expected positions.

12. Computer program, characterized in that it comprises instructions for implementing the method according to one of claims 1 to 11 when the program is executed by a processor.

13. Device for verifying the positions of fuel assemblies arranged within a nuclear reactor vessel, each fuel assembly comprising at least one reference element, the device comprising:

- a memory for storing captured images from a camera fdming at least the reference elements of at least a portion of the assemblies,

- a computer system cooperating with the memory, for denoising the captured images and locating the reference elements in the denoised images, characterized in that, as the camera films close-ups of only a portion of the assemblies in each image, and is moving along a predetermined path to obtain a sequence comprising consecutive images of all the assemblies, the computer system determines an area of interest in each captured image based on said predetermined path, the area of interest including one or more expected positions of reference elements, with a tolerance around said expected positions so as to also include the actual positions of the reference elements in the captured images, and that the computer system performs the denoising and determines the location of reference elements only in said area of interest, for each image.