The resonant inspection method is widely used in die-casting part manufacturers to detect defect parts in production. Common defects in die-casting parts are cracks and presence of porosity, which will cause the natural frequencies shifts of the parts. Depending on different size or position of the defect, in some cases these frequency shifts are very small, which has reduced the applicability of the resonant inspection method. Different from the resonant inspection method which only uses the natural frequencies information, machine learning technique can use the high-dimensional features of the whole data of transfer function, which can enhance the accuracy and robustness of the recognition of defects. In this paper, an application of machine learning on defect detection of die-casting steering wheel armature is presented. An integrated automatic testing machine is developed and is used in the production line of die-casting steering wheel armature for high volume 100% inspection. The frequency transfer function of the armature is obtained by a modal testing system and is used in the defect detection software written based on the machining learning algorithm. After the training of the algorithm based on data from tens of thousands of productions, a more than 90% defect detection rate has been achieved during the on-line production process.
 

On-line detection,defect detection,Transfer function,machine learning