Our project with Deep Digital Cornwall (DDC) on the Digitisation & Automation of Optical Petrography reaches the end of ERDF funding this month, and we would like to share its main outcomes. The first has been the setup of an automated ZEISS AxioImager optical microscope and the integration of its large-area scanning capability into our standard petrographic workflows. All in-house concrete sections are now scanned in three light paths (PPL – plane polarised light, XPL – cross polarised light and EPO – episcopic fluorescence). The images, although incredible to view, really make a positive impact on company productivity because of the development of a bespoke Java-based image viewing tool created by our DDC project lead, Laura Carter-Greaves. The Petrolab View (PLV) tool allows images to be loaded and viewed on any laptop or desktop in the office or even remotely. This allows for more flexible working opportunities and also breaks the bottleneck of requiring additional petrographic microscopes and all their ancillary parts as our petrography team builds. Remote access also allows for greater collaboration among colleagues, clients and, potentially, external consultants. PLV’s smooth scrolling, magnification and selection of light paths make assessment of thin sections much quicker than by standard optical petrography. The PLV tool can also capture areas of interest for reporting, adding a scale and watermark automatically to the snapshot, further saving time.
These are simple but effective benefits of the new PLV workflows. However, the addition of a point counting function really adds to the value of this project, as not only does it allow greater flexibility for staff to conduct modal analysis of samples without waiting for a mechanical counter to become available, but the results can also easily be reviewed and corrected if the wrong key is pressed or a point miss-identified. As an overlay for every point is recorded, points can easily be relocated and reviewed on screen by colleagues, thereby making the tool excellent for training junior staff and also for internal data validation and quality control.
The first half of this project has been focused on improving standard industry methods and allowing Petrolab to scale and improve the quality of our petrographic services. The focus of the second part of the project, in conjunction with the University of Exeter Mathematics department, is to explore and develop automated methods of analysis using machine learning (ML) and image analysis techniques. This cutting-edge project has been led by Dr Saptarshi Das, Senior lecturer in Mathematics, who has begun to look at a variety of different clustering image analysis tools and their ability to separate and measure key metrics from the section scans. We are close to validating the initial methods and hope to publish the findings in the coming year. This is certainly the tip of the iceberg for this application, and we hope to continue the ML research with a PhD collaboration.
Overall we are extremely pleased with the successful outcomes of this project and thank the DDC team for their strong support and encouragement throughout. We are very excited to continue this work and see digitisation at the forefront of the continuing development and improvement of our petrographic services.