Neural network based manufacturability evaluation of free form machining

Most CAD/CAM and computer-aided process planning systems manipulate all geometrical features on the part equally. In the area of free form machining, lack of efficient methodology for assessing the degree of manufacturing pretentiousness of free form features is still noticeable. Developing this methodology inside CAD/CAM systems brings the following benefits to the tool shop praxis: it minimizes the number of set-ups and tool changes and at the same time ensures the right sequence of machining strategies in order to achieve the best possible surface quality in the machining area. Based on this assessment, the CAD/CAM process will also be greatly simplified. When there are an increased number of non-prismatic and non-cylindrical features, this problem is even more exaggerated, and its solution cannot be found in the framework of analytical mathematics. This paper reports a neuro-fuzzy model that uses the concept of “feature manufacturability” to identify and recognize the degree of “pretentiousness-difficulty of machining”. The model is created by means of the construction of parametric fuzzy membership functions, based on neural networks learning process. This makes possible simultaneous evaluation of features complexity in a CAD model and manufacturing capability in an environment description model.