The use of neural networks in ultrasonic flaw detection

Ultrasound based inspection techniques are used extensively throughout industry for detection of flaws in engineering materials. The range and variety of imperfections encountered is large and critical assessment of location, size, orientation and type is often difficult. In addition, increasing quality requirements of new standards and codes of practice relating to fitness for purpose are placing higher demands on operators. The use of neural networks to support the operator in the initial evaluation and classification process offers advantages in terms of improvement in measurement accuracy and reduction in operator work load. This paper presents an evaluation of various types and configurations of neural networks developed for the purpose of assisting in accurate flaw detection in steel plates. The research presented was conducted using a wide range of samples including non-defective plates, side drilled holes, inclusions and porosity together with smooth and rough cracks. The results obtained indicate that significant benefits may be obtained from the techniques demonstrated.