JP4211092B2 - Automatic welding defect detection method in radiographic inspection - Google Patents

Automatic welding defect detection method in radiographic inspection Download PDF

Info

Publication number
JP4211092B2
JP4211092B2 JP24179898A JP24179898A JP4211092B2 JP 4211092 B2 JP4211092 B2 JP 4211092B2 JP 24179898 A JP24179898 A JP 24179898A JP 24179898 A JP24179898 A JP 24179898A JP 4211092 B2 JP4211092 B2 JP 4211092B2
Authority
JP
Japan
Prior art keywords
processing
defect
image
weld
defect detection
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Lifetime
Application number
JP24179898A
Other languages
Japanese (ja)
Other versions
JP2000076446A (en
Inventor
昇平 御園
俊則 生川
定男 出川
政行 栗原
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
IHI Corp
Original Assignee
IHI Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by IHI Corp filed Critical IHI Corp
Priority to JP24179898A priority Critical patent/JP4211092B2/en
Publication of JP2000076446A publication Critical patent/JP2000076446A/en
Application granted granted Critical
Publication of JP4211092B2 publication Critical patent/JP4211092B2/en
Anticipated expiration legal-status Critical
Expired - Lifetime legal-status Critical Current

Links

Images

Landscapes

  • Image Analysis (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
  • Image Processing (AREA)

Description

【0001】
【発明の属する技術分野】
本発明は、放射線透過より溶接欠陥を検出する溶接欠陥検出法に係り、特に放射線透過装置により観察される溶接欠陥を画像処理技術により自動検出する放射線透過検査における溶接欠陥自動検出法に関するものである。
【0002】
【従来の技術】
従来、配管同士などを突合せ溶接した後、その突合せ溶接部は、欠陥が発生していないか検査されている。
【0003】
この溶接欠陥を検出する方法としては、放射線透過装置でその溶接部に放射線を照射し、専門の検査員による目視観察によって溶接欠陥の有無及びそのグレード分けを評価する方法が知られている。
【0004】
また、この方法以外にも、放射線透過装置により溶接部に放射線を照射した画像を入力し、その入力画像を画像処理して溶接欠陥を検出する溶接欠陥検出方法について、数多く発表されている。
【0005】
【発明が解決しようとする課題】
しかしながら、検査員の目視による評価は、その結果において、個人差が生じ、また定性的な判断しかできないため、きめ細かなグレード分けが困難であった。さらに、溶接欠陥は、形状等の特徴により、溶接部の溶け込みが不足した場合などに発生する線状欠陥、溶接部内に気孔が発生した球状欠陥、及び溶接部から垂れが発生したたれ欠陥等に分類されるが、これらを識別し、総合的な評価をするには、溶接に関する専門知識が必要である。
【0006】
また、従来の画像処理による溶接欠陥検出方法は、放射線透過装置によって撮像される溶接部の画像は、コントラストが非常に小さいため、上述した線状欠陥、球状欠陥、及びたれ欠陥等の各種溶接欠陥を精度良く分離・識別することはできなかった。
【0007】
そこで、本発明の目的は、放射線透過装置によって放射線が照射された溶接部から、画像処理技術により各種欠陥を自動的に精度良く分離・識別して検出すると共に、それらのグレードを定量的に評価できる放射線透過検査における溶接欠陥自動検出法を提供することにある。
【0008】
【課題を解決するための手段】
上記課題を解決するために請求項1の発明は、突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像から、溶接部の輪郭線を抽出し、その輪郭線から正常部の溶接線を近似計算させ、その近似させた線と輪郭線とを重ね合せてずれた領域を抽出する凸状たれ欠陥検出処理と
を行う方法である。
【0010】
請求項の発明は、突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理と
を行う方法である。
【0011】
請求項の発明は、突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像を2段階しきい値処理して、欠陥候補領域を抽出すると共に溶接領域を抽出し、これら抽出領域をAND処理してノイズを除去し、溶接領域内に形成された球状欠陥を抽出する球状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像から、溶接部の輪郭線を抽出し、その輪郭線から正常部の溶接線を近似計算させ、その近似させた線と輪郭線とを重ね合せてずれた領域を抽出する凸状たれ欠陥検出処理
を行う方法である。
【0012】
請求項の発明は、突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像を2段階しきい値処理して、欠陥候補領域を抽出すると共に溶接領域を抽出し、これら抽出領域をAND処理してノイズを除去し、溶接領域内に形成された球状欠陥を抽出する球状欠陥検出処理と、
上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理
を行う方法である。
【0013】
請求項の発明は、線状欠陥検出処理と凸状たれ欠陥検出処理に加え、上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理を行う方法である。
【0014】
請求項の発明は、線状欠陥検出処理と球状欠陥検出処理と凸状たれ欠陥検出処理に加え、上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理を行う方法である。
【0015】
上記構成によれば、ノイズが多く低コントラストの画像でも、形状情報を用いた検出処理を、欠陥ごとに適用することにより、各種欠陥を精度良く分離・識別し、定量的な評価を自動で行うことができる。
【0016】
【発明の実施の形態】
次に、本発明の好適実施の形態を添付図面に基づいて詳述する。
【0017】
図1に本発明にかかる放射線透過検査における溶接欠陥自動検出法の処理フローを示す。
【0018】
図1に示すように、本発明にかかる放射線透過検査における溶接欠陥自動検出法は、放射線透過装置により溶接部に放射線を照射して撮像する画像入力処理10、その入力画像を正規化する正規化処理20、その正規化画像から溶接欠陥を検出する欠陥検出処理30,40,50,60、検出した溶接欠陥の等級を計算する欠陥等級計算処理70、その溶接欠陥の種類及び等級により総合的に評価する総合評価値の算出処理80の順に行われる。
【0019】
欠陥検出処理は、検出する欠陥の種類に応じて、線状欠陥検出処理、球状欠陥検出処理、凸状たれ欠陥検出処理、及び塊状たれ欠陥検出処理が行われる。
【0020】
これら各種欠陥の検出処理について、管同士の突合せ溶接部から溶接欠陥を検出することを例に詳述する。
【0021】
まず、線状欠陥検出処理30について図2〜図5を用いて説明する。
【0022】
図2に線状欠陥検出処理の処理フローを示す。
【0023】
図2に示すように、線状欠陥検出処理30は、正規化処理20後の画像を、放射線照射量の相違による濃淡差を補正する濃淡モフォロジ処理31し、線状欠陥候補領域を多重多段階スライス法32で抽出した後、線状欠陥候補領域を折れ線ベクトルで近似表現する折れ線ベクトル化処理33を行う。そして、それら折れ線ベクトルを、直線の方向および、位置関係を考慮して統合する。すなわち、線状欠陥は、飛び飛びの領域に離れているものの、大局的には1本の直線上にあることから、大局的探索法による線状欠陥抽出処理34することにより、検出される。
【0024】
線状欠陥領域の濃度分布は、一定値とは限らず、局所的に波をうつような分布をしている。そのため、2値化しきい値を高くするとバラバラに途切れ、低くすると多くのノイズを抽出してしまう。そこで、濃淡分布が不連続に変化する線状欠陥の候補領域を抽出するのに、多重多段階スライス処理を適用する。
【0025】
多重多段階スライス処理は、局所的な濃淡を繰り返す領域を抽出し、かつノイズの包含を避ける処理であり、その基本原理は、まず濃淡モフォロジ処理した画像に対して、非常に低い濃度しきい値T1 で2値化することによって、断片的ではあるが確実に欠陥の一部であると判定される部分を抽出する。次いで、そのしきい値よりも少し高い濃度しきい値範囲{T1 +(i−1)α <T<T1 +iα i=1,,,N}の部分を取り出し、それがこれまでに抽出した部分に隣接するか否かを調べ、隣接している場合には欠陥領域として延長登録する。この操作をN回繰り返す。さらに、以上のことを、低いしきい値T1 からM回繰り返し、これらを統合した部分を欠陥領域とする。このように、同じ処理をM回繰り返すことにより、ノイズを統合せずに、線状欠陥のみを良好に抽出することができる。
【0026】
この多重多段階スライス処理32を図3を用いてより具体的に説明する。
【0027】
実際の線状欠陥領域21とノイズ領域22が、図3上部の位置の濃度グラフに示したような濃度分布23にある時を例に説明する。
【0028】
▲1▼ 非常に低いしきい値T1 で対象画像を2値化する。これにより、線状欠陥領域の一部32aを抽出する。
【0029】
▲2▼ 次いで、対象画像に対してT1 よりも少し高い濃度しきい値範囲{T1 <T<T2 }の部分32bを抽出する。
【0030】
▲3▼ ▲2▼の各抽出部分32bの領域が、▲1▼の抽出部分野領域32aと近接しているか否かを調べ、近接している領域については、線状欠陥領域21aとして拡張登録する。この際、線状欠陥領域21aにある中央の抽出部分32cは、▲1▼の抽出部分32aの領域と近接していないため、拡張されず、同様にノイズ領域22の抽出部分32dは、▲1▼の抽出部分32aの領域と近接していないため、拡張登録されない。
【0031】
▲4▼ 対象画像に対して、よりしきい値の高い濃度のしきい値範囲{T2 <T<T3 }の部分32dを抽出する。
【0032】
▲5▼ ▲4▼の抽出部分32dの領域が、▲3▼の線状欠陥領域21aと近接しているか否かを調べ、近接している領域については、線状欠陥領域21bとして拡張登録する。また、線状欠陥領域21dにある抽出部分32dは、▲3▼の線状欠陥領域21aと近接していないため、拡張登録されず、同様にノイズ領域22の抽出部分32eは、▲3▼の線状欠陥抽出領域21aと近接していないため、拡張登録されない。
【0033】
▲6▼ 対象画像に対して、さらに高い濃度のしきい値範囲{T3 <T<T4 }の部分32fを抽出する。
【0034】
▲7▼ ▲6▼の抽出部分32fの領域が、▲5▼の線状欠陥領域21bと近接しているか否かを調べ、近接している領域については、線状欠陥領域21cとして拡張登録する。同様にノイズ領域22の抽出部分32gについては、▲5▼の線状欠陥抽出領域21bと近接していないため、拡張登録されない。
【0035】
▲8▼ ▲1▼〜▲7▼の処理を同じ濃度しきい値範囲で再び行うことで、▲5▼の線状欠陥領域に連結されなかった▲4▼の抽出部分32dが線状欠陥と近接しているため、拡張登録されて、線状欠陥領域21dとされる。また、ノイズである▲4▼の抽出部分32eは、近接していないため、拡張登録されない。
【0036】
▲9▼ ▲8▼で得られた拡張登録した線状欠陥領域21dの画像を新たな▲1▼の画像と見なして、さらに▲1▼〜▲7▼の処理を行って線状欠陥領域21eを抽出する。この際、ノイズ領域22の抽出部分32d,32e,32gは、線状欠陥領域21cと近接せず、かつ▲8▼の抽出部分21dと近接していないため、ノイズ領域32hとされる。
【0037】
この▲1▼〜▲9▼の例では、3回の拡張処理を3回繰り返しているが、上述したように、拡張回数N回、繰り返し回数M回として一般化できる。
【0038】
この時点では、コンピュータ内での線状欠陥候補領域は、画像の各画素ごとの処理しか行っていないので、隣接関係や幾何学的な形状に関する情報は何も得られていない。そこで、一連の線情報を全て折れ線ベクトルで近似表現する折れ線ベクトル化処理33を施し、端点座標、長さや傾きといった直線情報に記述する。すなわち、一つながりの曲線を複数の折れ線で近似することによって、その長さや各折れ線の傾きなどの幾何学的形状をコンピュータ内に記述する。これによって抽出した全ての線状欠陥候補群を記号として認識できる。この線状欠陥候補群を、線状欠陥候補領域セグメント群と呼ぶ。
【0039】
さらに次の処理として、この線状欠陥候補領域セグメント群の中で、線状欠陥を探索する大局的探索処理34を行う。
【0040】
この大局的探索処理34は、線状欠陥候補領域の抽出処理により得られた領域から、線状欠陥の連続性に着目し、真の線状欠陥を抽出する処理であり、図4に示すように、その探索領域を極力減らして処理効率を上げるために、真の線状欠陥であろうと推定される所定のしきい値以上の長さの直線セグメントを選択34aし、探索の起点とする。以後、この直線セグメントをベースセグメントと呼ぶ。
【0041】
そして、各ベースセグメントごとに、他の候補セグメントとのペアで直線を形成できるか否かを判定34bし、形成可能である場合には、図5(a)に示すように、最小2乗法によってそれらぺアのセグメントs0 ,s1 に近似する直線lを生成する。
【0042】
そして、図5(b)に示すように、この直線l上に存在する候補セグメントs2 を全て検出34cする。そして、セグメント群s0 ,s1 ,s2 が構成する全直線の長さがあるしきい値以上で、かつその全体の長さに占めるセグメントの比率が高ければ、図5(c)に示すように、それらのセグメントを結合して線状欠陥sであると判定34dする。その際、セグメントs,s2 同士があまりにも大きく(所定の長さより)離れている場合は、分断して、それぞれの直線群に対して欠陥であるかどうかの判定を行う。
【0043】
次に、球状欠陥検出処理40について図6〜図8を用いて説明する。
【0044】
球状欠陥検出処理は、球状欠陥は溶接部内に気孔が発生したものであることから、溶接領域内において、濃度差が比較的小さく、かつ面積の小さな領域を抽出する処理である。このため、単純2値化処理で球状欠陥を抽出する場合、2値化しきい値を低くすると、ノイズ領域を多く誤抽出し、逆に2値化しきい値を高くすると、ノイズは低減されるが、球状欠陥領域の一部分しか抽出されなくなる。そこで、2段階しきい値処理によって、球状欠陥候補領域を検出する。
【0045】
2段階しきい値処理は、まず低いしきい値で2値化することによって、ノイズを含む欠陥領域全体を抽出し、次に、ノイズを含まず球状欠陥の一部のみを抽出するような高いしきい値で2値化することによって、確実に球状欠陥である非常に明るい部分のみを抽出し、最後に、低いしきい値で2値化した領域の中で、高いしきい値で2値化した領域が含まれるものを抽出する処理である。図7に2段階しきい値処理の模式図を示す。
【0046】
図7に示すように、濃淡モフォロジ処理後の画像を、2段階しきい値処理することにより、低いしきい値で2値化した画像42aからは、球状欠陥42kを検出できるが、ノイズ42nも検出される。また、濃淡モフォロジ処理後の画像を、高いしきい値で2値化した画像42bからは、溶接部内の球状欠陥42kの一部のみが抽出される。すなわち、高いしきい値で2値化した画像42bからは、低いしきい値で2値化した領域の中で、高い濃度値を持たないものはノイズとして除去される。そして、これら2画像(領域)を合わせた画像42cから、重ならない領域はノイズ42nと見なして除去され、球状欠陥42kのみが抽出される。ただし、2段階しきい値処理によって、溶接領域以外の部分に、ノイズが抽出されることがある。このため、本発明にあっては、欠陥候補領域を抽出する明部抽出用の2段階しきい値処理の他に、暗部抽出用の2段階しきい値処理を行って溶接領域を抽出する。
【0047】
図6に本発明にかかる球状欠陥検出処理フローを示す。
【0048】
図6に示すように、球状欠陥検出処理は、正規化処理20後の画像を、濃淡モフォロジ処理41し、上述した方法と同様にして、明部抽出用の2段階しきい値処理による欠陥候補領域を抽出42する。他方、溶接領域全体を抽出する低いしきい値と溶接領域の一部のみを抽出する高いしきい値とで、暗部抽出用の2段階しきい値処理を行い、この2段階しきい値処理によって得られた画像をノイズ除去処理し、さらに溶接領域内の穴埋め処理を行って溶接領域を抽出43する。その後、これら欠陥候補領域と溶接領域の2画像(領域)をAND処理44した画像に対して、さらに、ノイズ除去処理45を行うことにより、実際の球状欠陥を検出する。図8に本発明の球状欠陥検出処理の模式図を示す。
【0049】
図8に示すように、明部抽出用の2段階しきい値処理により抽出された画像42eの欠陥候補領域には、球状欠陥43kの他にもノイズ43nも含まれている。また、暗部抽出用の2段階しきい値処理により抽出された画像43aの領域は、ノイズ43nを有していると共に欠陥領域の穴43hを有している。そして、ノイズ除去処理により、溶接領域43y以外の領域のノイズ43nが除去され、穴埋め処理により、欠陥領域の穴43hが穴埋めされ、溶接領域のみが抽出された画像43cが得られる。さらに、これら2画像(領域)42e,43cのAND処理44により、球状欠陥候補領域からノイズ43nが除去され、溶接領域内の球状欠陥のみが抽出された画像44aが得られる。
【0050】
次に、凸状たれ欠陥検出処理50について図9、図10を用いて説明する。
【0051】
図9に凸状たれ欠陥抽出処理フローに示す。
【0052】
凸状たれ欠陥は、溶接部から溶接金属が管の内側に垂れて溶接領域の内側輪郭線に凸状に現れる欠陥であるので、溶接領域と濃淡差がほとんどなく、濃度情報だけでは、検出することができない。そこで、管同士の突合せ溶接の場合、溶接部の輪郭線に対して管の溶接線を形成するための2次曲線近似を2段階で行い、管の内側境界線(輪郭線)から溶接部の輪郭線のずれ領域を抽出する処理を行い、凸状たれ欠陥を抽出する方法を発明した。
【0053】
図9に示すように、この凸状たれ欠陥検出処理40は、まず、正規化処理20した画像を、大きなフィルタサイズで濃淡モフォロジ処理51し、その濃淡モフォロジ処理した画像に対して、溶接領域全体を抽出する低いしきい値と溶接領域の一部のみを抽出する高いしきい値の2段階しきい値処理を適用して、溶接領域を抽出52する。この方法では、溶接欠陥の背景画像を用いないため、溶接された管の位置に影響されずに、溶接領域のみを良好に抽出することができる。
【0054】
そして、得られた溶接領域の内側境界線(輪郭線)をサンプリングし、点列データとして記述53した後、図10上部に示すように、最小2乗法による2次曲線近似処理54を行う。ただし、この2次曲線l0 は、凸状部の点も含めて計算されているので、正常部(溶接線)に相当する内側境界線(輪郭線)を正確には表していない。そこで、図10中央部に示すように、近似2次曲線から大きく離れたオリジナル点列データoを削除した後、再度、最小2乗法による2次曲線近似処理55を行う。これによって、正常部の内側境界線(輪郭線)lを忠実に表すことができる。
【0055】
最後に、図10下部に示すように、2回目の近似曲線lとオリジナル点列データoで囲まれる凸状領域tの長さL、高さW、及び面積Sがそれぞれ、しきい値以上であれば、凸状たれ欠陥として抽出する。
【0056】
次に、塊状たれ欠陥検出処理60について図11、図12を用いて説明する。
【0057】
塊状たれ欠陥は、溶接領域中央部に円形状に現れる欠陥で、周囲との濃淡差が比較的大きい特徴を持つ。そこで、塊状たれ欠陥抽出処理は、溶接領域内で溶接領域外との濃度差が大きな領域を抽出する処理を行う。
【0058】
図11に塊状たれ欠陥検出処理フローを示す。
【0059】
図11に示すように、まず、正規化処理20後の画像を、単純2値化処理61を行い、塊状たれ欠陥の候補領域を抽出62する。この段階では、溶接された管の外側領域や溶接領域以外のノイズも抽出してしまうため、この画像から面積の大きな領域(濃度差がほぼ一定の領域)を削除する。他方、正規化処理20後の画像を濃淡モフォロジ処理63した後、2段階しきい値処理によって溶接領域を抽出64する。その後、塊状たれ欠陥候補領域と溶接領域のAND処理65を行って溶接領域内の候補領域のみを抽出し、ノイズ除去処理66して塊状たれ欠陥を検出する。図12に本発明の塊状たれ欠陥検出処理の模式図を示す。
【0060】
図12に示すように、単純2値化処理により、管の外側領域61o及び溶接欠陥候補領域61kとの画像61aが抽出され、この抽出画像61aから面積の大きな領域が削除され、溶接欠陥候補領域61kのみの画像62aが抽出される。また、球状欠陥抽出処理と同様に、2段階しきい値処理により、溶接領域64yの画像64aが抽出される。そして、これら欠陥候補領域62aと溶接領域64yの2つの画像62a,64aのAND処理によって、溶接領域64y外のノイズ61nが除去されて溶接領域64y内の候補領域61kのみが抽出され、ノイズ除去処理によって溶接領域64a内のノイズ64nが除去され、塊状たれ欠陥61kが検出される。
【0061】
そして、このようにして検出された溶接欠陥は、各種欠陥ごとにその大きさ(等級)を求める欠陥等級計算が行われる。
【0062】
この等級計算は各種欠陥ごとに行われ、線状欠陥等級計算処理71は、線状欠陥の長さを計算し、球状欠陥等級計算処理72は、球状欠陥の面積を計算し、また、たれ欠陥等級計算処理73は、凸状たれ欠陥と塊状たれ欠陥の面積を計算する。
【0063】
さらに、この計算値をもとに溶接欠陥の総合評価値の算出処理80が行われる。
【0064】
総合評価値の算出処理80は、各溶接欠陥ごとに、欠陥の種類や大きさから総合的に、JIS規格などに基いてグレード分けを行う。
【0065】
以上説明したように、本発明は、溶接欠陥の検出処理として、線状欠陥検出処理、球状欠陥検出処理、凸状たれ欠陥検出処理、及び塊状たれ欠陥検出処理の4種類の処理を行うことで、これらを精度良く分類できると共に、それらのグレードを定量的に評価できる。
【0066】
また、本発明は、それぞれの欠陥検出処理がノイズを除去する作用を含むことから、溶接欠陥の画像が低コントラストであっても、正確に溶接欠陥を検出できる。
【0067】
尚、本実施の形態では、凸状たれ欠陥検出処理において、管同士の溶接部から欠陥領域を抽出するため、2次曲線近似を行ったが、平板同士の突合せ溶接部から欠陥領域を抽出する場合には直線近似を行うなど、溶接線の形状に応じた近似を行うことにより、凸状たれ欠陥を検出することができる。
【0068】
【発明の効果】
以上要するに本発明によれば、以下に示すような効果を発揮する。
【0069】
(1)放射線透過による溶接欠陥検査において、従来、目視で行っていたため、検査員による個人差が生じていたが、本発明により、線状欠陥、球状欠陥、凸状たれ欠陥、及び塊状たれ欠陥を自動的に精度良く分離・識別して検出でき、更にそれらのグレードを定量的に評価できる。
【0070】
(2)画像処理システムを構築することにより、専門の検査員でなくとも、溶接欠陥検査を行うことができるようになり、さらに工場における溶接ラインを自動化することができる。
【図面の簡単な説明】
【図1】本発明にかかる放射線透過検査における溶接欠陥自動検出法の流れ図である。
【図2】図1の線状欠陥検出処理の処理フローを示す図である。
【図3】図2の多重多段階スライス処理手順を示す図である。
【図4】図2の大局的探索処理の処理フローを示す図である。
【図5】図4の大局的探索処理の模式図である。
【図6】図1の球状欠陥検出処理の処理フローを示す図である。
【図7】図6の球状欠陥検出処理の2段階しきい値処理の模式図である。
【図8】図6の球状欠陥検出処理の模式図である。
【図9】図1の凸状たれ欠陥検出処理の処理フローを示す図である。
【図10】図9の凸状たれ欠陥検出処理の模式図である。
【図11】図1の塊状たれ欠陥検出処理の処理フローを示す図である。
【図12】図11の塊状たれ欠陥検出処理の模式図である。
【符号の説明】
10 画像入力処理
20 正規化処理
30 線状欠陥検出処理
40 球状欠陥検出処理
50 凸状たれ欠陥検出処理
60 塊状たれ欠陥検出処理
70 欠陥等級計算処理
80 溶接欠陥の総合評価値の算出処理
[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a welding defect detection method for detecting a welding defect by radiation transmission, and more particularly to a welding defect automatic detection method in a radiation transmission inspection in which a welding defect observed by a radiation transmission device is automatically detected by an image processing technique. .
[0002]
[Prior art]
Conventionally, after butt welding pipes and the like, the butt welded portion is inspected for defects.
[0003]
As a method for detecting this welding defect, a method is known in which the welded portion is irradiated with radiation by a radiation transmission device, and the presence or absence of the welding defect and its grade classification are evaluated by visual observation by a specialized inspector.
[0004]
In addition to this method, many welding defect detection methods have been disclosed in which an image obtained by irradiating a welded portion with radiation by a radiation transmission device is input, and the input image is subjected to image processing to detect a welding defect.
[0005]
[Problems to be solved by the invention]
However, the evaluation by the inspector's visual inspection has resulted in individual differences in the results, and since only qualitative judgment can be made, it has been difficult to make detailed grades. Furthermore, weld defects include linear defects that occur when the weld is insufficiently melted due to features such as shape, spherical defects that have pores in the weld, and dripping defects that have drooped from the weld. Although it is classified, expertise in welding is necessary to identify and comprehensively evaluate them.
[0006]
Further, in the conventional welding defect detection method using image processing, since the image of the welded portion imaged by the radiation transmission device has a very small contrast, various types of welding defects such as the above-described linear defects, spherical defects, and sagging defects are included. Could not be separated and identified with high accuracy.
[0007]
Accordingly, an object of the present invention is to automatically separate and identify various defects from a welded portion irradiated with radiation by a radiation transmission device automatically and accurately by image processing technology, and quantitatively evaluate their grades. An object of the present invention is to provide an automatic detection method for welding defects in radiographic inspection.
[0008]
[Means for Solving the Problems]
In order to solve the above-mentioned problems, the invention of claim 1 is directed to a welding defect automatic detection method in a radiographic inspection for inspecting a defect in a weld from an inspection image obtained by irradiating a butt weld with radiation.
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
The contour line of the welded part is extracted from the image after the above shade morphology processing, the weld line of the normal part is approximated from the contour line, and the shifted area is extracted by superimposing the approximated line and the contour line. And a convex defect detection process .
[0010]
The invention of claim 2 is an automatic welding defect detection method in a radiographic inspection for inspecting defects in a weld from an inspection image obtained by irradiating a butt weld with radiation.
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
A defect candidate area having a large density difference is extracted by binarization processing from the image after the normalization processing, the image is subjected to density morphology processing, and a welding region is extracted by two-stage threshold processing. This is a method of performing AND processing on the extraction region and performing block defect detection processing for extracting a region having a large density difference in the welding region.
[0011]
The invention of claim 3 is an automatic welding defect detection method in a radiographic inspection for inspecting a defect of a weld from an inspection image obtained by irradiating a butt weld with radiation.
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
The image after the density morphology processing is subjected to two-stage threshold processing to extract defect candidate regions and weld regions, and these extracted regions are AND-processed to remove noise and formed in the weld regions. Spherical defect detection processing for extracting spherical defects;
The contour line of the welded part is extracted from the image after the above shade morphology processing, the weld line of the normal part is approximated from the contour line, and the shifted area is extracted by superimposing the approximated line and the contour line. it is convex sauce defect detection process and a method of performing <br/> to.
[0012]
The invention of claim 4 is a welding defect automatic detection method in a radiographic inspection for inspecting defects in a weld from an inspection image obtained by irradiating a butt weld with radiation.
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
The image after the density morphology processing is subjected to two-stage threshold processing to extract defect candidate regions and weld regions, and these extracted regions are AND-processed to remove noise and formed in the weld regions. Spherical defect detection processing for extracting spherical defects;
A defect candidate area having a large density difference is extracted by binarization processing from the image after the normalization processing, the image is subjected to density morphology processing, and a welding region is extracted by two-stage threshold processing. the extraction region by aND operation, a method of performing <br/> the bulk sauce defect detection process of extracting a larger area of the density difference in the welding area.
[0013]
In addition to the linear defect detection process and the convex defect detection process, the invention of claim 5 extracts a defect candidate area having a large density difference by binarization from the normalized image, and A method for performing block defect detection processing in which a density morphology process is performed on an image, a welding region is extracted by performing two-stage threshold processing, and these extraction regions are ANDed to extract a region having a large density difference in the welding region. It is.
[0014]
In the invention of claim 6 , in addition to the linear defect detection process, the spherical defect detection process, and the convex defect detection process, a defect candidate area having a large density difference is binarized from the image after the normalization process. Extraction and density morphology processing of the image, two-step threshold processing to extract the welded area, AND these extracted areas AND processing, to extract a large density difference area in the welded area This is a method of performing detection processing.
[0015]
According to the above configuration, even in a noisy and low-contrast image, by applying detection processing using shape information for each defect, various defects are accurately separated and identified, and quantitative evaluation is automatically performed. be able to.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
Next, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.
[0017]
FIG. 1 shows a processing flow of a welding defect automatic detection method in a radiographic inspection according to the present invention.
[0018]
As shown in FIG. 1, an automatic welding defect detection method in a radiographic inspection according to the present invention includes an image input process 10 for irradiating and imaging a welded portion with a radiographic apparatus, and normalization for normalizing the input image Processing 20, defect detection processing 30, 40, 50, 60 for detecting a welding defect from the normalized image, defect grade calculation processing 70 for calculating the grade of the detected welding defect, and comprehensively based on the type and grade of the welding defect It is performed in the order of the comprehensive evaluation value calculation process 80 to be evaluated.
[0019]
In the defect detection process, a linear defect detection process, a spherical defect detection process, a convex defect detection process, and a block defect detection process are performed according to the type of defect to be detected.
[0020]
The detection processing of these various defects will be described in detail by taking, as an example, detection of a welding defect from a butt weld between pipes.
[0021]
First, the linear defect detection process 30 will be described with reference to FIGS.
[0022]
FIG. 2 shows a process flow of the linear defect detection process.
[0023]
As shown in FIG. 2, the linear defect detection process 30 performs a density morphology process 31 that corrects a density difference due to a difference in radiation dose on the image after the normalization process 20, and multi-stages the linear defect candidate areas. After extraction by the slicing method 32, a polygonal line vectorization process 33 for approximating the linear defect candidate area with a polygonal line vector is performed. Then, these broken line vectors are integrated in consideration of the direction of the straight line and the positional relationship. In other words, the linear defect is detected by performing the linear defect extraction process 34 by the global search method since it is on a single straight line although it is separated from the jumping region.
[0024]
The concentration distribution of the linear defect region is not limited to a constant value, but has a distribution that is locally waved. Therefore, if the binarization threshold value is increased, it will break apart, and if it is lowered, a lot of noise will be extracted. Therefore, multiple multi-stage slice processing is applied to extract candidate areas of linear defects whose density distribution changes discontinuously.
[0025]
Multiple multi-step slicing is a process that extracts regions that repeat local shading and avoids the inclusion of noise. Its basic principle is that a very low density threshold is first applied to an image that has been subjected to shading morphology processing. By binarizing with T 1 , a portion that is determined to be a part of a defect although it is fragmented is extracted. Next, a portion of the density threshold range {T 1 + (i−1) α <T <T 1 + iα i = 1,..., N} that is slightly higher than the threshold is taken out and extracted so far. It is checked whether or not it is adjacent to the portion, and if it is adjacent, it is registered as an extension as a defective area. This operation is repeated N times. Further, the above, repeated M times from low threshold T 1, and these integrated part of the defective area. In this way, by repeating the same process M times, it is possible to satisfactorily extract linear defects without integrating noise.
[0026]
The multiple multi-stage slice processing 32 will be described more specifically with reference to FIG.
[0027]
A case where the actual linear defect area 21 and the noise area 22 are in the density distribution 23 as shown in the density graph at the upper part of FIG. 3 will be described as an example.
[0028]
( 1 ) The target image is binarized with a very low threshold value T 1 . Thereby, a part 32a of the linear defect region is extracted.
[0029]
{Circle around (2)} Next, a portion 32b of the density threshold range {T 1 <T <T 2 } that is slightly higher than T 1 is extracted for the target image.
[0030]
(3) It is checked whether or not the area of each extraction portion 32b in (2) is close to the extraction part field area 32a in (1), and the adjacent area is expanded and registered as a linear defect area 21a. To do. At this time, the central extracted portion 32c in the linear defect region 21a is not expanded because it is not close to the region of the extracted portion 32a of (1), and similarly, the extracted portion 32d of the noise region 22 is Since it is not close to the area of the extraction portion 32a of ▼, it is not registered for extension.
[0031]
{Circle around (4)} A portion 32d of a threshold range {T 2 <T <T 3 } having a higher threshold is extracted from the target image.
[0032]
(5) Check whether the area of the extraction part 32d in (4) is close to the linear defect area 21a in (3), and the adjacent area is expanded and registered as a linear defect area 21b. . The extracted portion 32d in the linear defect region 21d is not close to the linear defect region 21a of (3) and is therefore not registered for expansion. Similarly, the extracted portion 32e of the noise region 22 is Since it is not close to the linear defect extraction area 21a, the extended registration is not performed.
[0033]
{Circle around (6)} A portion 32f of a threshold range {T 3 <T <T 4 } having a higher density is extracted from the target image.
[0034]
It is checked whether the area of the extraction part 32f in (7) (6) is close to the linear defect area 21b in (5), and the adjacent area is expanded and registered as a linear defect area 21c. . Similarly, the extracted portion 32g of the noise region 22 is not registered in an expanded manner because it is not close to the linear defect extraction region 21b of (5).
[0035]
By performing the processing of (8) (1) to (7) again within the same density threshold range, the extraction portion 32d of (4) that is not connected to the linear defect region of (5) is regarded as a linear defect. Since they are close to each other, they are expanded and registered as a linear defect region 21d. In addition, the extraction part 32e of (4) that is noise is not adjacent and is not registered in an extended manner.
[0036]
{Circle around (9)} The image of the linearly registered defect area 21d obtained in [8] is regarded as a new image of [1], and the processes of [1] to [7] are further performed to perform the linear defect area 21e. To extract. At this time, the extraction portions 32d, 32e, and 32g of the noise region 22 are not close to the linear defect region 21c and are not close to the extraction portion 21d of (8), and thus are set as the noise region 32h.
[0037]
In the examples {circle around (1)} to {circle around (9)}, the three expansion processes are repeated three times, but can be generalized as the number of expansions N times and the number of repetitions M as described above.
[0038]
At this point, since the linear defect candidate area in the computer is only processed for each pixel of the image, no information about the adjacency relationship or geometric shape is obtained. Therefore, a line vectorization process 33 for approximating all the series of line information with a line vector is performed and described in straight line information such as end point coordinates, length and inclination. That is, by approximating a continuous curve with a plurality of polygonal lines, a geometric shape such as the length and inclination of each polygonal line is described in the computer. Thus, all the extracted linear defect candidate groups can be recognized as symbols. This linear defect candidate group is called a linear defect candidate area segment group.
[0039]
Further, as a next process, a global search process 34 for searching for a linear defect in the linear defect candidate area segment group is performed.
[0040]
This global search process 34 is a process for extracting a true linear defect from the area obtained by the extraction process of the linear defect candidate area, focusing on the continuity of the linear defect, as shown in FIG. In addition, in order to reduce the search area as much as possible and increase the processing efficiency, a straight line segment having a length equal to or longer than a predetermined threshold estimated to be a true linear defect is selected 34a as a search starting point. Hereinafter, this straight line segment is referred to as a base segment.
[0041]
Then, for each base segment, it is determined whether or not a straight line can be formed in pairs with other candidate segments, and if it can be formed, as shown in FIG. 5A, the least square method is used. A straight line l approximating the segments s 0 and s 1 of the pair is generated.
[0042]
Then, as shown in FIG. 5B, all candidate segments s 2 existing on the straight line l are detected 34c. If the lengths of all the straight lines formed by the segment groups s 0 , s 1 , s 2 are equal to or greater than a certain threshold value and the ratio of the segments to the total length is high, the result is shown in FIG. As described above, it is determined that the segment is a linear defect s by combining the segments. At this time, if the segments s and s 2 are too large (separate from a predetermined length), they are divided and a determination is made as to whether or not each line group is defective.
[0043]
Next, the spherical defect detection process 40 will be described with reference to FIGS.
[0044]
The spherical defect detection process is a process for extracting a region having a relatively small concentration difference and a small area in the welded region because the spherical defect is a hole in the welded portion. For this reason, when a spherical defect is extracted by simple binarization processing, if the binarization threshold value is lowered, many noise areas are erroneously extracted. Conversely, if the binarization threshold value is increased, noise is reduced. Only a part of the spherical defect area is extracted. Therefore, a spherical defect candidate region is detected by a two-step threshold process.
[0045]
In the two-stage threshold processing, binarization is first performed at a low threshold value, thereby extracting the entire defect area including noise, and then extracting only a part of the spherical defect including no noise. By binarizing with a threshold value, only a very bright portion that is surely a spherical defect is extracted, and finally, in a region binarized with a low threshold value, it is binary with a high threshold value. This is a process of extracting a region including the converted area. FIG. 7 shows a schematic diagram of the two-stage threshold processing.
[0046]
As shown in FIG. 7, a spherical defect 42k can be detected from an image 42a binarized with a low threshold by performing two-step threshold processing on the image after the density morphology processing, but noise 42n is also detected. Detected. Further, only a part of the spherical defect 42k in the welded portion is extracted from the image 42b obtained by binarizing the image after the density morphology processing with a high threshold value. That is, from the image 42b binarized with a high threshold value, an area that does not have a high density value among the areas binarized with a low threshold value is removed as noise. Then, from the image 42c obtained by combining these two images (regions), the non-overlapping region is considered as noise 42n, and only the spherical defect 42k is extracted. However, noise may be extracted in a portion other than the welding region by the two-stage threshold processing. For this reason, in the present invention, in addition to the two-stage threshold process for extracting the bright part for extracting the defect candidate area, the two-stage threshold process for extracting the dark part is performed to extract the welding region.
[0047]
FIG. 6 shows a spherical defect detection processing flow according to the present invention.
[0048]
As shown in FIG. 6, in the spherical defect detection process, the image after the normalization process 20 is subjected to the density morphological process 41, and in the same manner as described above, defect candidates are obtained by the two-stage threshold process for extracting the bright part. An area is extracted 42. On the other hand, two-stage threshold processing for dark portion extraction is performed with a low threshold for extracting the entire welding region and a high threshold for extracting only a part of the welding region. The obtained image is subjected to noise removal processing, and further, a welding region is extracted 43 by performing hole filling processing in the welding region. Thereafter, an actual spherical defect is detected by further performing noise removal processing 45 on an image obtained by AND processing 44 of these two images (regions) of the defect candidate region and the welding region. FIG. 8 shows a schematic diagram of the spherical defect detection processing of the present invention.
[0049]
As shown in FIG. 8, the defect candidate area of the image 42e extracted by the two-step threshold process for extracting the bright part includes noise 43n in addition to the spherical defect 43k. The region of the image 43a extracted by the two-stage threshold processing for dark portion extraction has a noise 43n and a hole 43h of a defective region. Then, the noise 43n is removed from the region other than the welded region 43y by the noise removing process, and the hole 43h in the defective region is filled by the hole filling process, and an image 43c in which only the welded region is extracted is obtained. Further, the AND process 44 of these two images (regions) 42e and 43c removes the noise 43n from the spherical defect candidate region, thereby obtaining an image 44a in which only the spherical defect in the welding region is extracted.
[0050]
Next, the convex defect detection processing 50 will be described with reference to FIGS.
[0051]
FIG. 9 shows a convex defect extraction process flow.
[0052]
Convex sag defect is a defect in which the weld metal hangs down from the weld to the inside of the tube and appears in a convex shape on the inner contour line of the welded region, so there is almost no difference in density from the welded region, and it is detected only by the density information. I can't. Therefore, in the case of butt welding between pipes, a quadratic curve approximation for forming the weld line of the pipe with respect to the outline of the weld is performed in two stages, and the weld boundary is determined from the inner boundary line (contour line) of the pipe. Invented a method of extracting a convex defect by performing a process of extracting a contour shift area.
[0053]
As shown in FIG. 9, the convex defect detection process 40 first performs a density morphology process 51 on the normalized image 20 with a large filter size, and the entire welded region is subjected to the density morphology process. The welding region is extracted 52 by applying a two-step threshold processing of a low threshold value for extracting a high threshold value and a high threshold value for extracting only a part of the welding region. In this method, since the background image of the weld defect is not used, only the weld region can be extracted well without being affected by the position of the welded pipe.
[0054]
Then, after sampling the inner boundary line (contour line) of the obtained welded region and describing 53 as point sequence data, a quadratic curve approximation process 54 by the least square method is performed as shown in the upper part of FIG. However, since this quadratic curve l 0 is calculated including the points of the convex portion, it does not accurately represent the inner boundary line (contour line) corresponding to the normal part (weld line). Therefore, as shown in the center of FIG. 10, after deleting the original point sequence data o far away from the approximate quadratic curve, the quadratic curve approximation process 55 by the least square method is performed again. Thus, the inner boundary line (contour line) 1 of the normal part can be faithfully represented.
[0055]
Finally, as shown in the lower part of FIG. 10, the length L, the height W, and the area S of the convex region t surrounded by the second approximate curve l and the original point sequence data o are each greater than or equal to the threshold value. If there is, it is extracted as a convex sag defect.
[0056]
Next, the block defect detection process 60 will be described with reference to FIGS.
[0057]
The lump defect is a defect that appears in a circular shape at the center of the welding region, and has a characteristic that the difference in density from the surroundings is relatively large. Therefore, the block defect extraction process performs a process of extracting a region having a large density difference between the outside of the welding region and the outside of the welding region.
[0058]
FIG. 11 shows a block defect detection processing flow.
[0059]
As shown in FIG. 11, first, the binarized defect candidate region is extracted 62 by performing a simple binarization process 61 on the image after the normalization process 20. At this stage, noise outside the welded tube and other areas than the welded region are also extracted, so a region with a large area (region with a substantially constant density difference) is deleted from this image. On the other hand, after the image after the normalization process 20 is subjected to the density morphology process 63, a welding region is extracted 64 by a two-stage threshold process. Thereafter, an AND process 65 is performed between the block defect candidate area and the weld area to extract only the candidate area in the weld area, and a noise removal process 66 is performed to detect the block defect. FIG. 12 shows a schematic diagram of the lump defect detection processing of the present invention.
[0060]
As shown in FIG. 12, the image 61a of the pipe outer region 61o and the welding defect candidate region 61k is extracted by the simple binarization process, the region having a large area is deleted from the extracted image 61a, and the welding defect candidate region is obtained. An image 62a of only 61k is extracted. Similarly to the spherical defect extraction process, the image 64a of the weld region 64y is extracted by the two-step threshold process. The noise 61n outside the welding region 64y is removed by AND processing of the two images 62a and 64a of the defect candidate region 62a and the welding region 64y, and only the candidate region 61k within the welding region 64y is extracted, and the noise removal processing is performed. As a result, the noise 64n in the welding region 64a is removed, and the block defect 61k is detected.
[0061]
And the defect grade calculation which calculates | requires the magnitude | size (grade) of the welding defect detected in this way for every defect is performed.
[0062]
This grade calculation is performed for each type of defect, the linear defect grade calculation process 71 calculates the length of the linear defect, the spherical defect grade calculation process 72 calculates the area of the spherical defect, and the sagging defect. The grade calculation processing 73 calculates the areas of the convex sag defect and the block sag defect.
[0063]
Further, based on this calculated value, a calculation process 80 of a comprehensive evaluation value of welding defects is performed.
[0064]
In the comprehensive evaluation value calculation process 80, each welding defect is graded based on the type and size of the defect based on the JIS standard.
[0065]
As described above, the present invention performs four types of processing, that is, linear defect detection processing, spherical defect detection processing, convex defect detection processing, and block defect detection processing as welding defect detection processing. These can be classified with high accuracy and their grades can be quantitatively evaluated.
[0066]
Moreover, since each defect detection process includes the effect | action which removes noise, this invention can detect a welding defect correctly even if the image of a welding defect is low contrast.
[0067]
In this embodiment, quadratic curve approximation is performed in order to extract the defect region from the welded portion between the tubes in the convex sag defect detection process, but the defect region is extracted from the butt welded portion between the flat plates. In some cases, by performing approximation according to the shape of the weld line, for example, by performing linear approximation, a convex defect can be detected.
[0068]
【The invention's effect】
In short, according to the present invention, the following effects are exhibited.
[0069]
(1) In the welding defect inspection by radiation transmission, since it was conventionally performed by visual inspection, individual differences by the inspector occurred, but according to the present invention, a linear defect, a spherical defect, a convex defect, and a block defect. Can be automatically separated and identified with high accuracy, and their grades can be quantitatively evaluated.
[0070]
(2) By constructing an image processing system, it becomes possible to perform a weld defect inspection without being a specialized inspector, and further to automate a welding line in a factory.
[Brief description of the drawings]
FIG. 1 is a flowchart of a welding defect automatic detection method in a radiographic inspection according to the present invention.
FIG. 2 is a diagram showing a processing flow of the linear defect detection processing of FIG. 1;
FIG. 3 is a diagram showing a multiplex multi-stage slice processing procedure of FIG. 2;
4 is a diagram showing a processing flow of the global search processing of FIG. 2;
FIG. 5 is a schematic diagram of the global search process of FIG. 4;
6 is a diagram showing a processing flow of the spherical defect detection processing of FIG. 1; FIG.
7 is a schematic diagram of a two-step threshold process of the spherical defect detection process of FIG. 6. FIG.
FIG. 8 is a schematic diagram of the spherical defect detection process of FIG. 6;
9 is a diagram showing a processing flow of convex defect detection processing in FIG. 1; FIG.
10 is a schematic diagram of the convex deflection defect detection process of FIG. 9; FIG.
FIG. 11 is a diagram showing a processing flow of the block defect detection processing of FIG. 1;
12 is a schematic diagram of the block defect detection process of FIG. 11. FIG.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 10 Image input process 20 Normalization process 30 Linear defect detection process 40 Spherical defect detection process 50 Convex defect detection process 60 Block defect detection process 70 Defect grade calculation process 80 Calculation process of comprehensive evaluation value of weld defect

Claims (6)

突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像から、溶接部の輪郭線を抽出し、その輪郭線から正常部の溶接線を近似計算させ、その近似させた線と輪郭線とを重ね合せてずれた領域を抽出する凸状たれ欠陥検出処理と
を行うことを特徴とする放射線透過検査における溶接欠陥自動検出法。
In the welding defect automatic detection method in the radiographic inspection for inspecting the defect of the weld from the inspection image obtained by irradiating the butt weld with radiation,
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
The contour line of the welded part is extracted from the image after the above shade morphology processing, the weld line of the normal part is approximated from the contour line, and the shifted area is extracted by superimposing the approximated line and the contour line. A method for automatically detecting a welding defect in a radiographic inspection, characterized by performing a convex defect detection process .
突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理と
を行うことを特徴とする放射線透過検査における溶接欠陥自動検出法。
In the welding defect automatic detection method in the radiographic inspection for inspecting the defect of the weld from the inspection image obtained by irradiating the butt weld with radiation,
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
A defect candidate area having a large density difference is extracted by binarization processing from the image after the normalization processing, the image is subjected to density morphology processing, and a welding region is extracted by two-stage threshold processing. An automatic welding defect detection method in a radiographic inspection characterized by performing AND processing on an extraction region and performing block defect detection processing for extracting a region having a large density difference in the welding region.
突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像を2段階しきい値処理して、欠陥候補領域を抽出すると共に溶接領域を抽出し、これら抽出領域をAND処理してノイズを除去し、溶接領域内に形成された球状欠陥を抽出する球状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像から、溶接部の輪郭線を抽出し、その輪郭線から正常部の溶接線を近似計算させ、その近似させた線と輪郭線とを重ね合せてずれた領域を抽出する凸状たれ欠陥検出処理
を行うことを特徴とする放射線透過検査における溶接欠陥自動検出法。
In the welding defect automatic detection method in the radiographic inspection for inspecting the defect of the weld from the inspection image obtained by irradiating the butt weld with radiation,
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
The image after the density morphology processing is subjected to two-stage threshold processing to extract defect candidate regions and weld regions, and these extracted regions are AND-processed to remove noise and formed in the weld regions. Spherical defect detection processing for extracting spherical defects;
The contour line of the welded part is extracted from the image after the above shade morphology processing, the weld line of the normal part is approximated from the contour line, and the shifted area is extracted by superimposing the approximated line and the contour line. welding automatic defect detection in radiographic testing, which comprises carrying out convex sauce defect detection process and the <br/> to.
突合せ溶接部に放射線を照射して得られた検査画像から溶接部の欠陥を検査するための放射線透過検査における溶接欠陥自動検出法において、
上記検査画像を正規化処理した後、濃淡モフォロジ処理し、その処理後の画像から、多重多段階スライス処理にて線状欠陥を抽出する線状欠陥検出処理と、
上記濃淡モフォロジ処理後の画像を2段階しきい値処理して、欠陥候補領域を抽出すると共に溶接領域を抽出し、これら抽出領域をAND処理してノイズを除去し、溶接領域内に形成された球状欠陥を抽出する球状欠陥検出処理と、
上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理
を行うことを特徴とする放射線透過検査における溶接欠陥自動検出法。
In the welding defect automatic detection method in the radiographic inspection for inspecting the defect of the weld from the inspection image obtained by irradiating the butt weld with radiation,
After normalization processing of the inspection image, density morphology processing, and from the processed image, linear defect detection processing for extracting linear defects in multiple multi-stage slice processing,
The image after the density morphology processing is subjected to two-stage threshold processing to extract defect candidate regions and weld regions, and these extracted regions are AND-processed to remove noise and formed in the weld regions. Spherical defect detection processing for extracting spherical defects;
A defect candidate area having a large density difference is extracted by binarization processing from the image after the normalization processing, the image is subjected to density morphology processing, and a welding region is extracted by two-stage threshold processing. the extraction region by aND operation, welding defects automatic detection in radiographic testing, wherein the bulk sauce defect detection process of extracting a larger area to make a <br/> density difference in the weld region.
線状欠陥検出処理と凸状たれ欠陥検出処理に加え、上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理を行う請求項記載の放射線透過検査における溶接欠陥自動検出法。In addition to the linear defect detection process and the convex defect detection process, the defect candidate area having a large density difference is extracted from the image after the normalization process by the binarization process, and the image is subjected to the density morphology process. 2-step thresholding to extract the weld region, these extraction region by aND operation, radiographic testing according to claim 1, wherein performing the bulk sauce defect detection process of extracting a larger area of the density difference in the weld region Automatic detection of welding defects. 線状欠陥検出処理と球状欠陥検出処理と凸状たれ欠陥検出処理に加え、上記正規化処理後の画像から、2値化処理にて濃度差の大きな欠陥候補領域を抽出すると共に、その画像を濃淡モフォロジ処理し、2段階しきい値処理して溶接領域を抽出し、これら抽出領域をAND処理して、溶接領域内で濃度差の大きな領域を抽出する塊状たれ欠陥検出処理を行う請求項記載の放射線透過検査における溶接欠陥自動検出法。In addition to the linear defect detection process, the spherical defect detection process, and the convex defect detection process, a defect candidate area having a large density difference is extracted from the image after the normalization process by the binarization process. gray morphology processing, two-step thresholding to extract the weld area, these extraction regions aND processing, claim 3 for bulk sauce defect detection process of extracting a larger area of the density difference in the weld region Automatic welding defect detection method in the described radiographic inspection.
JP24179898A 1998-08-27 1998-08-27 Automatic welding defect detection method in radiographic inspection Expired - Lifetime JP4211092B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP24179898A JP4211092B2 (en) 1998-08-27 1998-08-27 Automatic welding defect detection method in radiographic inspection

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP24179898A JP4211092B2 (en) 1998-08-27 1998-08-27 Automatic welding defect detection method in radiographic inspection

Publications (2)

Publication Number Publication Date
JP2000076446A JP2000076446A (en) 2000-03-14
JP4211092B2 true JP4211092B2 (en) 2009-01-21

Family

ID=17079679

Family Applications (1)

Application Number Title Priority Date Filing Date
JP24179898A Expired - Lifetime JP4211092B2 (en) 1998-08-27 1998-08-27 Automatic welding defect detection method in radiographic inspection

Country Status (1)

Country Link
JP (1) JP4211092B2 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102175700A (en) * 2011-01-20 2011-09-07 山东大学 Method for detecting welding seam segmentation and defects of digital X-ray images

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4618502B2 (en) * 2005-07-11 2011-01-26 株式会社Ihi Fluorescence flaw detector and fluorescent flaw detection method
JP4618501B2 (en) * 2005-07-11 2011-01-26 株式会社Ihi Fluorescence flaw detector and fluorescent flaw detection method
JP4664778B2 (en) * 2005-09-08 2011-04-06 ヤンマー株式会社 Chestnut quality inspection method
JP4702952B2 (en) * 2006-06-01 2011-06-15 大日本スクリーン製造株式会社 Unevenness inspection method, unevenness inspection apparatus, and program
JP4702953B2 (en) * 2006-06-01 2011-06-15 大日本スクリーン製造株式会社 Unevenness inspection method, unevenness inspection apparatus, and program
JP5166909B2 (en) * 2008-02-25 2013-03-21 三菱重工業株式会社 Inspection device and inspection method
US8369481B2 (en) * 2009-06-08 2013-02-05 Ishida Co., Ltd. X-ray inspection device
JP5943722B2 (en) * 2012-06-08 2016-07-05 三菱重工業株式会社 Defect determination apparatus, radiation imaging system, and defect determination method
CN108876771B (en) * 2018-06-04 2022-05-20 广东工业大学 Undercut welding defect detection method
JP7251158B2 (en) * 2019-01-16 2023-04-04 Tdk株式会社 Evaluation method of raw material alloy for rare earth magnet, apparatus for evaluating raw material alloy for rare earth magnet, and manufacturing method of rare earth magnet

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102175700A (en) * 2011-01-20 2011-09-07 山东大学 Method for detecting welding seam segmentation and defects of digital X-ray images
CN102175700B (en) * 2011-01-20 2012-07-25 山东大学 Method for detecting welding seam segmentation and defects of digital X-ray images

Also Published As

Publication number Publication date
JP2000076446A (en) 2000-03-14

Similar Documents

Publication Publication Date Title
CN113362326B (en) Method and device for detecting defects of welding spots of battery
US6876445B2 (en) Method for analyzing defect data and inspection apparatus and review system
CN108009591B (en) Contact network key component identification method based on deep learning
JP4211092B2 (en) Automatic welding defect detection method in radiographic inspection
JP4189061B2 (en) Crack detection method for concrete surface
JP2009145285A (en) Defect detecting method and defect detecting apparatus
CN111986195B (en) Appearance defect detection method and system
Diaz et al. Rust detection using image processing via Matlab
KR100687811B1 (en) Method and apparatus for detecting defects of the welded part
JP3572750B2 (en) Automatic evaluation method for concrete defects
Alam et al. An algorithm to detect and identify defects of industrial pipes using image processing
JPH0961138A (en) Crack extraction apparatus
JP4318776B2 (en) Nonuniformity inspection method and apparatus
JPH1048152A (en) Method and apparatus for detecting fine linear defect
KR20160097651A (en) Apparatus and Method for Testing Pattern of Sample using Validity Image Processing Technique, and computer-readable recording medium with program therefor
CN111738991A (en) Method for creating digital ray detection model of weld defects
Sundaram et al. Welding defects extraction for radiographic images using C-means segmentation method
Abd Halim et al. The geometrical feature of weld defect in assessing digital radiographic image
JP2006337167A (en) Inspection method of low-contrast flaw under cyclic noise, and inspection method of low-contrast flaw under repeated pattern
CN116071306A (en) Machine vision detection method for riveting defect of rolling bearing retainer
JP3126634B2 (en) Appearance inspection method
JP2003247954A (en) Defect detection method for round body circumference
Da Silva et al. The state of the art of weld seam radiographic testing: Part ll-pattern recognition
JP2001033225A (en) Image processing and inspecting method and image processing and inspecting device
JP7372687B2 (en) Automatic inspection equipment, automatic pipe inner surface inspection equipment, and automatic appearance inspection equipment for three-dimensional structural products

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20050714

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20080722

A521 Request for written amendment filed

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20080909

TRDD Decision of grant or rejection written
A01 Written decision to grant a patent or to grant a registration (utility model)

Free format text: JAPANESE INTERMEDIATE CODE: A01

Effective date: 20081007

A01 Written decision to grant a patent or to grant a registration (utility model)

Free format text: JAPANESE INTERMEDIATE CODE: A01

A61 First payment of annual fees (during grant procedure)

Free format text: JAPANESE INTERMEDIATE CODE: A61

Effective date: 20081020

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20111107

Year of fee payment: 3

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20111107

Year of fee payment: 3

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20121107

Year of fee payment: 4

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20131107

Year of fee payment: 5

EXPY Cancellation because of completion of term