Additive Manufacturing Parameter Optimization with Automated Post-printing Flaw Detection Using Convolutional Neural Networks

In additive manufacturing, the printing quality and errors are inevitable nowadays. The quality errors such as not extruding at start of the print, not sticking to the bed, stringing or oozing, layer shifting, layer separation and stops extruding mid print can lead to complete wastage of material and time. Detecting such defects while printing the piece will help eradicating the wastage of material and saves lots of valuable time. Providing a proper checkpoints and critical design identification where the defects are highly vulnerable can help us perform corrective measures in the early stages of printing. Here we present our findings on a novel approach based on visual pattern mining using volumetric elemental pixels popularly known as Voxels. The proposed finding provides an accelerated process monitoring and detection of printing failure conditions—by the method of classifying every layer of the printing 3D model into critical and normal layers using advanced deep learning pattern mining approach with convolutional neural networks and automatic choice of critical checkpoints from the classification, to calculate error deviation between the Voxel image of critical 3D printed layer with actual image of same layer from the semi-finished part. Integration of a camera in 3D printer, Voxel separation and processing, pattern mining, and deep machine learning provides the above-proposed system which results in high test accuracy >93% on unknown raw models. Images of parts are taken at various stages of the printing process according to the part geometry, and Voxel images are extracted from >20k 3D models. A deep learning method, convolutional neural networks (CNN), is proposed to classify the parts into either ‘normal’ or ‘critical’ category. Parts using PLA and FDM materials were printed to demonstrate the proposed framework. We demonstrate that this methodology precisely and unambiguously detects the print failure in most cases and stops the print for manual corrective measure.