Genetic algorithms based parameter optimisation using statistical analysis I-kaz for groove defect classification

Abstract

Accurate signal acquisition is needed due to its implication on the structure integrity status and can lead to major structural failure due to immature signal analysis. The defect on the steel structure is usually associated with a loss of ferromagnetic material on the corrosion region and can be estimated by the magnitude of the Magnetic Flux Leakage (MFL) signal influenced by the depth, width and length of the defect. The challenge for researchers in determining a reliable defect characterisation are, first, the unstructured data with a lot of noise due to stray magnetic field, second, the MFL sensor selection and positioning, third, the parameter optimisation to enhance defect characterisation and reduce latency in postprocessing analysis. As a result, a Genetic Algorithm (GA) combines with the Integrated Kurtosis-based Algorithm for the Znotch filter (I-kazTM) technique for estimating MFL input parameters for defect due to material thinning is proposed. This technique produces a 3D graphic form presenting a degree of data scattering quantified by the coefficient of I-kaz (Z∞) This research aims to study the relationship between sensor acquisition speed and depth of defect using Z∞ using the multilevel signal decomposition technique as an output. The reliability analysis of the experiment is based on statistical verification such as ANOVA and R-square analysis. A complete MFL scanning system is developed using hall sensors and neodymium magnets integrated with a microcontroller as an instrument. The experimental setup comprises 12 Hall effect sensors positioned in an array of 10 cm in width attached on the linear guide for constant scanning speed on the selected setting. The workpiece consists of 10.0 mm, thickness carbon steel SAPH S45C with the designated defects depth of 1.0 mm, 3.0 mm and 6.0 mm to emulate the variability of defect sizes with multiple speed settings of 22.0 mms-1, 44.5 mms-1 and 65.5 mms-1 for each of the thickness. I-kaz multilevel signal decomposition shows the Z∞ increased significantly as the groove and acquisition speed became higher. In total, nine sets of MFL inspection experiments were conducted with various combinations of acquisition speed (20 mms-1 - 70 mms-1), designated depth of defect (1.0 mm to 6.0 mm) and Z∞ as result (0.05 to 0.400). The Response Surface Methodology (RSM) technique is then deployed using historical input from the MFL experiment by inputting speed and depth of defect as input parameters and Z∞ values as an output parameter to generate an equation model of fitness function between the input and output. The fitness function is then used in the GA to estimate the optimum speed and allowable defect depth that can be detected using the developed MFL inspection system by analysing the Z∞ value on each of the experiment runs. The parameter optimisation can be used to produce reliable output data and significant defect characterisation. The result shows the MFL inspection system provide a significant contribution in defect identification and decision-making from an array of hall effect sensor into a single actual value. This system also comes out with a new fitness model using RSM by indicating critical components in MFL inspection using acquisition speed, size of the defect and Z∞ values through historical experimental data. The methodology proves that good resolution of data can be produced with optimised parameters using a single value Z∞ as an output.