Development of a deep learning machining feature recognition network for recognition of four pilot machining features

Over the past few decades, several methods have been introduced for the recognition of machining features from design files. These methods entail considerable effort to infer the characteristics of machining features from the design and manufacturing information commonly expressed in different and incompatible formats on the geometry, topography, dimensions, tolerances, and surface finishes. The competence of these traditional methods in implementing their assignment is still challenging. The deep learning approach as an advanced facility in artificial intelligence is considered as a promising substitute. This approach has already been successfully and efficiently employed in different facets of social life. Its use in manufacturing engineering and industries is considered as a breakthrough, especially in some disciplines such as process planning, and machining feature recognition is in its embryonic phase. The input data in the present methods is the output information of a computer-aided design system. This is commonly faced with some limitations, the most serious ones of which are the loss of data from the design file occurring due to the geometric interference of features and slow extraction of features due to the extensive and excessive information existing in the design files. In the present study, a deep learning-based system named MFR-Net has been developed for the recognition of machining features from the images of workpieces. In addition to CAD systems, other tools such as the camera images of workpieces can also be employed to enter the input. The MFR-Net can also identify other relevant information needed for machining, including the position of the features, dimensions, and various symbols such as numbers, decimal points, and positive and negative signs employed in tolerances, parentheses, etc.