The Customer is a leading manufacturer of drilling equipment for the petroleum industry.
The Customer supplied drilling equipment and maintenance services to the petroleum production industry leaders and needed to monitor the condition of the drill bits they offered, providing drill bit wear analysis and timely defects detection. Physical inspection of the drill bits was time-consuming and expensive, so the Customer needed to optimize the condition monitoring process. The Customer chose to employ 3D cameras and needed to develop image recognition software for analyzing the drill bit images.
ScienceSoft delivered an application that uses machine learning and visual recognition algorithms to detect drill bit defects in the images captured by the cameras and provide recommendations on required drill bit replacement and maintenance.
Once launched, the application preprocesses an image and analyzes it to recognize blades and single out individual cutters, as well as to detect blade surface and cutter wear.
The application employs an object detection neural network relying on the Hough Circle Transform method to identify cutters. A simple convolutional neural network, a primary algorithm applied for image recognition, is used to perform cutter state classification, and a separate Mask R-CNN, an algorithm for object instance segmentation, is applied to perform blades’ surface segmentation and detect blade surface defects.
The application displays the results of the analysis as wear percentage and recommends an optimal date for drill bit replacement.
The Customer got the possibility to optimize drilling equipment condition monitoring and achieve timely detection of emerging drill bit defects. With the knowledge of drill bit replacement dates, the Customer got the chance to streamline their inventory management and lower inventory holding costs. The application also allowed the Customer to reduce drill bit inspection time, reducing downtime and lowering maintenance costs for their clients.
Technologies and Tools
TensorFlow, Keras, OpenCV, Scikit-learn, NumPy.