Typical Defect Classification And Damage Identification Of The Attachment Welds Of The Oil Pipeline

The traditional artificial detection, with many defects, can not satisfy the requirement of industrial production more and more. With the quick development of computer and image processing technology, welding defects detection gradually transits from artificial detection to computer intelligence recognition.The paper is based on the actual project demand and takes the pipe welding X-ray detection film as the research object, using intellectual means such as neural networks and expert systems etc, to realize the assessment automatically. After the accurate extraction of defect feature parameters, this paper attempts to build up a complete welding image recognition system based on the combination of neural networks and expert systems.The image recognition system is mainly composed of image processing module and classification and identification module. Image processing module mainly includes the following:noisy reducing, image enhancement, edge detection and image segmentation. The main contents of classification and identification module are as follows:a Self Organizing Feature Neural Network and a standard instrumental expert system. Parameters related to Self Organizing Feature Neural Network such as Weighting Matrix and Threshold Vector can be acquired by learning,achieving an input vector clustering effect.The traditional expert system is waken up automatically by a certain mechanism only when Self Organizing Feature Neural Network is in error with the recognition, then ensures the running stability of the recognition system. Aimed at the information stored in a database, the inference engine in the Expert System selects knowledge useful to current welds from KB, and performs reasoning by means of questions without no relevant information found. Proved by the experiments and simulation, the neural network is capable to recognize the current flaw sample and the defects of its own. But when new defects appear or there are no remarkable features with the defects, it's not surprising to see that Clustering of Neural Network in early period ends in failure. Given this situation, it is possible to recognize the corresponding defects with great accuracy and obtain the corresponding levels of welds with the help of expert system.This paper builds up the corresponding welding defects recognition system on VC++ platform, and the ideal processing results are achieved after realizing recognition of present defects.