SnapScan

About SnapScan

a. SanpScan a technology platform developed in conjunction with the Bits and Drilling Technology group. Its purpose is to decrease the T.C.O of Bits while improving performance and gaining market share. This is accomplished by embedding CA in the bit Engineering Change Request process to utilize the CA data to determine design requirements. To achieve the same,the CA platform consists of five integrated verticals :
i. The digital acquisition system with Mobile, Robotics and Data Matrix system capturing high volume data with precision.
ii. The AI empowered digital analytics uses images video for dull grading bits and generating repair recommendations.
iii. The 360 integrated view with Design, BOM, Maintenance, Run and Geology systems for each and every bit at cutter level.
iv. Big data analytics incorporating damage zone, design and cutter comparisons analysis for millions of cutters
v. AI driven optimization will be used to generate bit recommendations, cutter selection and design recommendations.
b. Overview: For cutter detection based on the edge device, the video of the blades taken @ 120 fps or another suitable frame rate is broken down into individual image frames. Each image is preprocessed and passed through multiple convolutional filters for removing noise and equalizing and normalizing image properties such as brightness, contrasts etc. Further they are aligned and rotated to make sure the algorithm is translational and rotationally invariant. Our patent claim includes using a small convolutional filter to predict object categories and offsets in bounding box locations, using separate predictors for different aspect ratio detections, and applying these filters to multiple feature maps to perform detection at multiple scales. With these modifications, we can achieve high-accuracy using relatively low resolution input, further increasing detection speed. The neural network architecture is further optimized for speed by reducing the hyperparameters need for tuning by introducing depth wise separable convolutions and linear bottleneck layers inspired from mobilenet v2 architecture. High speed tracking using kernelized correlation filters are used to track cutters identified in each frame of the video to maintain its unique identity. Such identity of each cutter in a video is sorted and aligned with prior knowledge from the design records. Further to embed the lighter compute version of the model to mobile device, we perform knowledge distillation for model compression by teaching a smaller network, step by step, exactly what to do using a bigger already trained network. The ‘soft labels’ refer to the output feature maps by the bigger network after every convolution layer. The smaller network is then trained to learn the exact behavior of the bigger network by trying to replicate it’s outputs at every level. The details of the deep neural network architecture are listed below.