Abstract:

Aiming at the difficulty of grade quantitative classification caused by the fuzziness of metal magnetic memory （MMM） characteristic parameters among different weld grades， a quantitative classification model based on particle swarm optimization fuzzy c-means clustering （FCM） is proposed. The fatigue tensile test was carried out on the Q235 steel weld specimen prefabricated by an incomplete penetration defect. The MMM signal was detected by the TSC-5M-32 MMM instrument. The three-dimensional composite characteristic parameter vector was extracted from the experimental data. At the same time， the MMM test results were compared with the X-ray test results to provide a reference. Considering that the initial clustering center of the FCM algorithm is determined randomly， it is easy to fall into the local optimum， and the artificial setting of weight m leads to low clustering accuracy. Particle swarm optimization （PSO） algorithm with global search and high efficiency is introduced to optimize the initial clustering center and weight m of the FCM algorithm. The modified reciprocal formula of the FCM objective function is used as the fitness function of the PSO algorithm， and the sample individuals and weight m are encoded as particles. The speed and position of the particle are updated to obtain the global optimal cluster center， and m is converged to the optimal solution. The quantitative MMM classification model based on the FCM clustering center and m optimized by PSO is established for different weld defect grades. The results show that the classification accuracy of the model is 97.93%， which provides a new idea for the quantitative identification of weld defect levels and evaluation of equipment safety.

Key words: metal magnetic memory, weld defect grades, fuzzy C-means, particle swarm optimization, cluster center