44 n MACHINE VISION April 2023 www.drivesncontrols.com been spent servicing more efficiently, while ensuring better spare parts availability and lower maintenance costs. Crucially, it means hugely improved uptime for our customers.” Creating a vision system that could deliver the accuracy the company required was a huge technical challenge. “Human vision tracking is something many of us take for granted, but even the most advanced camera is still vastly inferior to the human eye,” Reamsbottom explains. “Accurately reading a moving target like a barcode was something we always considered to be a huge technological barrier to overcome – particularly in varying light conditions – which is why we were amazed at the extraordinary performance of the upgraded camera. “We had already been working with B&R for many years, so when we discovered that they offered a camera system that could do what we needed, it seemed a logical step to take, as it fits seamlessly into the existing Powerlink communication system.” n The new vision system reads laser-etched 2D barcodes to track and trace the laundry bags Old machine vision hardware doesn’t need to be scrapped It is estimated that between 2017 and 2020, around 270,000 machine vision systems were installed at sites around the world. It is impossible to know how many of these are still performing as they were intended. But in a field where technology advances rapidly, systems that were cutting-edge a few years ago can quickly become outdated. A common problem is the lack of interoperability between different hardware components – or between hardware and software. Some software systems are vendor-specific, requiring users to buy new cameras and other sensors to apply the latest AI technologies. If users have already invested considerable amounts in machine vision equipment, do they really want to rip it out and start again? A joint venture between the Hong Kong based motors and actuators manufacturer Johnson Electric and the Israeli autonomous AI specialist Cortica has come up with an alternative to retrofitting. They have formed a visual inspection software company called Lean AI that uses patented deep-learning algorithms to automate the process of building an AI inspection models. If the existing vision equipment is acceptable in terms of image quality and illumination, then software is probably the main problem. Lean AI offers “equipment-agnostic” software, that can operate with hardware from different suppliers rather than being exclusive to one, allowing users to bring new capacity to their existing investments. Using unsupervised learning processes similar to those of the human brain, the software requires only non-labelled production data and no manual tagging. Lean AI is already putting its theory into practice at a powder metal company in Canada. The AI algorithms are linked to cameras and computers already in the plant, saving time, money and risk. Zohar Kantor, Lean AI’s vicepresident of sales, argues that if only a small fraction of existing machine vision systems are no longer performing the tasks they were originally designed for, there is a huge opportunity to retrofit equipment-agnostic software to give the hardware new life. n Many plants contain cameras, lighting and other visual inspection hardware that is no longer fit for purpose. But there is now an alternative to ripping it out and installing new hardware. AI-based software can breathe new life into existing machine vision hardware in some installations. Lean AI’s software can help to extend the operating lives of existing machine vision installations
RkJQdWJsaXNoZXIy MjQ0NzM=