Movement, pressure, temperature, humidity, sound frequency — sensors that are now integrated into our daily lives collect an endless stream of data about the way we interact with products and their environments.
From autonomous vehicles to health monitoring devices, the ever-growing amount of smart devices and information generated is becoming challenging to manage and more expensive to process.
“This massive amount of data needs to be stored and analyzed, and as a result, real-time processing is critical,” says Afshin Rahimi, an assistant professor of mechanical and aerospace engineering. “We are examining new techniques to accelerate the process.”
One of which is using gateway devices to analyze the data with deep learning models. This is called edge computing, and 90 per cent of industrial enterprises will be using it by 2022, according to a report by business consultants Frost & Sullivan.
The name edge is in reference to applying a deep learning model to analyze the data at the edge of a framework — where data is acquired — differing from cloud computing, which conducts the analysis on a remote server — where data is usually warehoused.
“Edge devices are especially promising for accelerating deep learning algorithms due to their low-power budget and high efficiency,” he adds. “It essentially brings data storage closer to the location where it is needed to improve response times and save bandwidth.”
Dr. Rahimi and his team are developing and applying computer vision and deep learning-based models using edge computing to analyze video recorded from manufacturing floors for process monitoring and proactive efficiency improvement.
The $90,000 project is funded by a Natural Sciences and Engineering Research Council of Canada (NSERC) Alliance grant and Ontario Centres of Excellence (OCE) Voucher for Innovation and Productivity (VIP) and industrial partner IFIVEO Canada Inc. (i-5O), a Silicon Valley-based start-up with operations worldwide, including Windsor.
For its remote production monitoring system, i-50 currently uses a cloud-based architecture to analyze video feed gathered from the shop floors of its clients, which are mostly large Fortune 500 manufacturers in North America and Asia.
“The primary issue with this approach is that the minimum Internet bandwidth required for streaming video is 3.5 Mbps and many of i-5O’s clients have fluctuating bandwidth that causes reliability issues,” Rahimi says. “Moreover, it is very costly to deploy deep learning models on the cloud, and it costs the company a significant amount per day per client location, cutting into gross margins.”
Khizer Hayat, i-5O’s chief innovation officer, says the partnership with Dr. Rahimi's team has been a key asset for the company. He points to ground-breaking algorithms developed by UWindsor PhD candidate, Pouria Modaresi, that helped the company improve its data accuracy while enhancing the client experience through better production insights.
“We're anticipating this project will produce significant performance improvements to our system architecture and have already discussed the new technology with customers in North America and Asia who are eager to test it out,” Hayat says. “We've been working with Dr. Rahimi's lab since 2019 and his team has produced some excellent innovations for i-5O.”
i-5O plans on testing the prototype with its customers in Canada, followed by the United States and Japan.