Conventional machine vision technology remains popular in the manufacturing factory, due to its proven repeatability, reliability, and stability. But business evolution suggests this may not be enough.
However, the emergence of deep learning technologies opens the possibility of expanded capabilities and flexibility, leading to more cost efficiency and higher production yield. Deep learning technologies offer so much potential that deep learning-based machine vision techniques in smart manufacturing will see a CGAR of 20% between 2017 and 2023, with a revenue that will reach US$34 billion by 2023, according to ABI Research, a market-foresight advisory firm providing strategic guidance on the most compelling transformative technologies.
Manufacturers are on the constant search to upgrade their production yields and workflow efficiency. Conventional machine vision is easy to implement but is limited in its capabilities. Current solutions that are widely deployed in quality control, safety inspection, predictive maintenance, and industrial monitoring rely on pre-programmed rules and criteria, supporting limited ranges of functions. Deep learning-based machine vision, however, is highly flexible due to its ability to be trained and improved using a new set of factory data, enabling manufacturers to incorporate updates and upgrade quickly.
“This is in part driven by the democratization of deep learning capabilities. The emergence of various open source Artificial Intelligence (AI) frameworks, such as TensorFlow, Caffe2, and MXNet lowers the barrier to entry for the adoption of deep learning-based machine vision,” said Lian Jye Su, a Principal Analyst at ABI Research. “These AI frameworks can be deployed using on-premise data centre infrastructure and a number of software packages from AI companies. In the past, the choice of machine vision solutions was limited to a handful of companies that performed relatively simple image processing operations. With deep learning-based machine vision, manufacturers can opt to develop their own deep learning-based machine vision systems without the worry of vendor lock-in.”
In addition to cameras, deep learning-based machine vision can also incorporate data collected from various sensors, including LiDAR, radar, ultrasound, and magnetic field sensors. The rich set of data will provide further insight into other aspects of production processes. As compared to conventional machine vision which can only detect product defects and quality issues which can be defined by humans, deep learning algorithms deployed for machine vision can go even further. These algorithms can pick up unexpected product abnormalities or defects, providing flexibility and valuable insights to manufacturers.
To implement deep learning-based machine vision technology, manufacturers are encouraged to work with a wide range of vendors, including industrial cloud platform, camera and sensor suppliers, and public cloud vendors. Deep learning-based machine vision requires a robust cloud platform that will enable condition-based monitoring, sensor data collection, and analytics. Unlike conventional machine vision which relies on line-by-line coding, deep learning-based machine vision models can be deployed by users without significant coding experience, as these models undergo unsupervised learning based on data gathered.
“Manufacturers are still opening up to adopting AI capabilities into their workflow. Deep learning-based machine vision will serve as the right catalyst to move the needle, as the potential is enormous. Startups that start off as deep learning-based machine vision solution providers are also starting to enable big data processing, process optimization, and yield analytics on their platform,” concluded Su.