In stage 1, the thermography image of rotating elements is first converted to a greyscale image using the Grey-scale conversion method. Then, contrast adjustment is performed through linear transformation followed by noise reduction and image smoothing using a median filter. In stage 2, the modified local binary pattern (MLBP) method is applied to detect the edges of the fault area using the local maxima average method. In stage 3, the proposed algorithm classifies the machine condition into three categories namely healthy condition, defect starts to arise and defect detection based on pixel intensity value in the thermal image.
This paper proposes a novel three-stage infrared thermography-based image segmentation method for defect detection in rotating machine elements. These three stages are, image pre-processing (stage 1), image segmentation by a MLBP (stage 2) and classification of various defect conditions (stage 3).
Five performance parameters namely Accuracy, Jaccard similarity index, Dice similarity coefficient, Sensitivity and Specificity are applied for the quantitative analysis of the proposed new MLBP method. Additionally, the output yield of the MLBP method is compared with seven existing bench-mark edge-based image segmentation methods, i.e. Fuzzy integrated with improved local binary pattern (Fuzzy-ILBP), linear transformation and improved hypersmoothing local binary pattern (LTIHLBP), local binary pattern (LBP), Canny, Sobel, Laplace of Gaussian (LoG) and Otsu-Canny. From the results, it is evident that the proposed method yields the highest accuracy of 98.85%, followed by Fuzzy-ILBP with an accuracy level of 98.70%.
Although the MLBP method successfully detects the most defective regions in the processed image, additional optimization is still required to improve the detection rate potential challenges include the initial cost of implementation, the need for regular maintenance of thermal imaging equipment and the requirement for skilled personnel to interpret the results.
The method can be scaled to handle big datasets and multiple machines simultaneously, making it suitable for large-scale industrial operations.
The no-contact, no-invasive and early detection capability of the proposed method gives higher economic growth and profits, higher productivity and cleaner and sustainable production. The proposed approach helps in achieving sustainable manufacturing by optimizing the production, by saving the time and by reduction the pollution and so on. These factors make the industry capable of an efficient production system and maintain the required product quality.
The method introduces an MLBP-based segmentation method with modifications specifically designed for rotating machinery thermal images, which surpasses seven alternative edge-based approaches in accuracy and performance. With accuracy reaching 98.85%, the technique showcases improved defect detection ability, providing a valuable tool for predictive maintenance and early fault diagnosis in industrial applications.
