Evolutionary Computing Firefly Algorithm Assisted Fuzzy C-Means Clustering Model for Steel Strip Surface Crack Detection
The exponential rise in the demands of quality manufacturing and cost-effective production systems have motivated academia-industries to develop more efficient and robust crack detection systems. The diverse morphological characteristics of the surface crack region confines efficiency of major existing crack-detection models. Especially the static threshold based crack detection models are limited with non-linear crack growth on target surface. Additionally, region growing concepts too require significant computation to yield satisfactory performance. On the other hand, clustering based crack detection methods have gained widespread attention; however, no significant effort has been made so far to detect non-linear crack regions on steel strip surface. Considering this as motive, in this paper a highly robust and novel Evolutionary Computing (EC) model named Firefly Algorithm based Fuzzy-C-Means (FA-FCM) model has been developed for steel-strip surface crack detection. Unlike classical clustering based approach, FA-FCM incorporates better clustering by obtaining optimal centroid information and reducing intra-cluster distance that eventually leads better clustering efficiency. To achieve better performance, intensity equalization followed by two-dimensional median filtering is performed over the target image. The efficient use of intensity equalization, median filtering and FA-FCM based clustering has enabled optimal crack detection and localization. The overall proposed crack detection model has been developed using MATLAB 2017a tool and has been assessed for performance over the real-time data collected from Jindal South West (JSW) Steels Ltd. The performance comparison in terms of both visual assessment as well as statistical investigation has revealed that the proposed FA-FCM based method exhibits better than the classical standard threshold based crack detection technique. The ease of implementation, computational efficiency and higher accuracy enable proposed model to be used for automatic real-time steel strip crack detection purposes.