Publication: Tren rayı muayenesi amaçlı manyetik kaçak akı sistem tasarımı ve yapay sinir ağı ile kusur karakterizasyonu
Abstract
Manyetik Kaçak Akı (MKA) yönteminde, test edilen ferromanyetik materyal içerisindeki akı yoğunluğunun saturasyona yakın oluşturulması gereklidir. Ancak uygun olmayan boyutlarda tasarlanan MKA sistemleri, manyetik köprünün kutup bölgelerinde yüksek akı yoğunlukları oluşturarak, istenmeyen kaçak akıların artmasına yol açmaktadır. Bu durum kusur sinyali üzerinde olumsuz etkiler oluşturmaktadır. Bu olumsuz etkilerin giderilmesi için akı yoğunluklarına göre sistem boyut optimizasyonu ile üretilen bir MKA sistem tasarımı gereklidir. Ayrıca bu alandaki diğer bir zorluk, kusur sinyalinden kusur özelliklerinin çıkartılması aşamasında tüm kusur kombinasyonları için çözüm sağlayan bir ilişkinin kurulmasıdır. Bu çalışmada tren rayı muayenesi için MKA sistem tasarımı ve kusur karakterizasyonu işlemleri gerçekleştirilmiştir. Optimum boyutlara sahip MKA sistemi, kaçak akılar ve ferromanyetik materyalin doğrusal olmayan davranışını temsil eden bir Manyetik Eşdeğer Devre (MED) modeli kullanılarak tasarlanmıştır. Tasarlanan MKA sistemi Finite Element Analizi (FEA) simülasyon ortamında oluşturularak farklı kusur kombinasyonları taşıyan tren rayı numuneleri üzerinde test işlemleri gerçekleştirilmiştir. FEA çalışmalarında ezilme ve çatlak tipi kusur örnekleri için kusur sinyalleri üretilerek kusur nitelik verileri çıkartılmıştır. Bu nitelik verileri minimum noktalar arası mesafe (L), maksimum-minimum noktalar arası fark (Tpp) ve alan (A) parametreleridir. FEA ortamında üretilen modeller esas alınarak aynı özellikleri taşıyan deneysel çalışmalar yapılmıştır. Elde edilen kusur sinyallerinden kusur nitelik verileri üretilerek tasarlanan bir Yapay Sinir Ağı (YSA) ile kusur karakterizasyonu gerçekleştirilmiştir. Tasarlanan YSA, FEA çalışmalarından üretilen kusur verileri referans alınarak eğitilmiştir. Eğitilen YSA, deneysel kusur testlerinde üretilen nitelik verilerinin girişe uygulanmasına karşılık çıkışında derinlik bilgisini üretmektedir. Deneysel çalışmalardan elde edilen sonuçlara göre YSA, kusur derinliği tahmininde minimum %93.6 başarı göstermiştir.
In the Magnetic Flux Leakage (MFL) method, the flux density in the ferromagnetic material under test should be close to saturation. However, MFL systems designed in inappropriate dimensions create high flux densities in the polar regions of the magnetic yoke, leading to an increase in unwanted leakage fluxes. This situation has adverse effects on the defect signal. In order to overcome these negative effects, a MFL system design produced by system size optimisation according to flux densities is required. In addition, another challenge in this area is to establish a relationship that provides a solution for all defect combinations during the defect feature extraction phase from the defect signal. In this study, MFL system design and defect characterisation processes for train track inspection are carried out. The optimum sized MFL system is designed using a Magnetic Equivalent Circuit (MEC) model representing the nonlinear behaviour of ferromagnetic material and leakage fluxes. The designed MFL system was created in the Finite Element Analysis (FEA) simulation environment and test procedures were carried out on train rail samples carrying different defect combinations. In FEA studies, defect signals were generated for crush and crack type defect samples and defect attribute data were extracted. These attribute data are minimum inter-point distance (L), maximum-minimum inter-point difference (Tpp) and area (A) parameters. Based on the models produced in the FEA environment, experimental studies with the same characteristics were carried out. Defect attribute data were generated from the obtained defect signals and defect characterisation was performed with a designed Artificial Neural Network (ANN). The designed ANN was trained with reference to the defect data generated from FEA studies. The trained ANN produces depth information at its output in response to the application of the attribute data produced in the experimental defect tests to the input. According to the results obtained from the experimental studies, the ANN showed a minimum success rate of 93.6% in defect depth prediction.
In the Magnetic Flux Leakage (MFL) method, the flux density in the ferromagnetic material under test should be close to saturation. However, MFL systems designed in inappropriate dimensions create high flux densities in the polar regions of the magnetic yoke, leading to an increase in unwanted leakage fluxes. This situation has adverse effects on the defect signal. In order to overcome these negative effects, a MFL system design produced by system size optimisation according to flux densities is required. In addition, another challenge in this area is to establish a relationship that provides a solution for all defect combinations during the defect feature extraction phase from the defect signal. In this study, MFL system design and defect characterisation processes for train track inspection are carried out. The optimum sized MFL system is designed using a Magnetic Equivalent Circuit (MEC) model representing the nonlinear behaviour of ferromagnetic material and leakage fluxes. The designed MFL system was created in the Finite Element Analysis (FEA) simulation environment and test procedures were carried out on train rail samples carrying different defect combinations. In FEA studies, defect signals were generated for crush and crack type defect samples and defect attribute data were extracted. These attribute data are minimum inter-point distance (L), maximum-minimum inter-point difference (Tpp) and area (A) parameters. Based on the models produced in the FEA environment, experimental studies with the same characteristics were carried out. Defect attribute data were generated from the obtained defect signals and defect characterisation was performed with a designed Artificial Neural Network (ANN). The designed ANN was trained with reference to the defect data generated from FEA studies. The trained ANN produces depth information at its output in response to the application of the attribute data produced in the experimental defect tests to the input. According to the results obtained from the experimental studies, the ANN showed a minimum success rate of 93.6% in defect depth prediction.
Description
Keywords
Artificial Neural Network, Defect characterization, Electric engineering, Electronics engineering, Elektrik mühendisliği, Elektromanyetik tahribatsız test, Elektronik mühendisliği, Kusur karakterizasyonu, Magnetic equivalent circuit, Magnetic Flux Leakage method, Manyetik eşdeğer devre, Manyetik Kaçak Akı yöntemi, Sistem Tasarımı Electromagnetic non-destructive testing, System Design, Testler ve ölçümler, Tests and measurements, Train rail, Tren rayı, Yapay Sinir Ağı
