Computer Vision ushers in a new era of manufacturing AI

The digital transformation in manufacturing is accelerating. This means that companies that make things must move faster to innovate their products and transform the way those products are delivered from design to procurement to manufacturing to sustainment. The good news? They don't have to go at it alone. A 2020 Deloitte and MAPI Study finds ecosystems can create a competitive edge for manufacturers facing ongoing disruption.¹ Today, we'll dive into a component of this ecosystem, Computer Vision.

Computer Vision is one of the many Artificial Intelligence solutions that will continue to transform manufacturing. Computer vision has been around for almost 50 years now; however, it is increasingly clear that the technology is ready for prime time. A 2019 Forrester survey found that 64% of global senior business purchase influencers say that computer vision will be very or extremely important to their firm in the coming year, and 58% said that their firm is implementing, planning to implement, or interested in implementing computer vision in the coming year.² It "allows machines to extract meaningful information from image and, now, even video data by using sophisticated machine learning models.³" Imagine a world where anything the human eye can see and brain can process, a computer can. Now go beyond that, as Computer Vision can reach past the natural human vision spectrum (e.g., lidar, radar, ultraviolet, thermal).

There are many different computer vision techniques ranging from image classification, which involves classifying images into predefined categories based on the actual content of the image to video analytics, which applies computer vision and spatio-temporal analyses to automatically identify significant information contained in video streams (a series of images or pixels), including the ability for human activity and posture to be detected and classified as different motions.

Computer vision technology is versatile and can be adapted for industries in different ways. In the supply chain and operations space, we've broken down a few use cases:

1. Quality Inspection – perhaps the most traditional use case for vision, this application began with detecting regions of interest along with the temporal states and AI-driven activity detection.  The analysis in this case is that of a single frame and use cases range from surface defects to dimensional tolerancing. The benefits here include a reduction in quality escapes, less rework, and a reduction in non-value-add inspection time from the front-line.
2. Factory Synchronization and Dynamic Scheduling – uses cameras mounted throughout a facility or loT to track material and product movement in real-time. This information is synced back to other manufacturing systems (e.g., ERP and MES) to dynamically schedule and allocate resources. Improvements from this use case include improved agility, reduction in inventory, and higher OEE.
3. Human-to-Machine (cobots) and Machine-to-Machine Collaboration (robots) – the human will never be replaced in the manufacturing facility, but we do know that augmenting them will result in better outcomes. As cobots become more commonplace and we ask robots to turn from a single player to a team-based approach, advanced computer vision techniques will be leveraged to keep people safe and optimize the way a team of robots knocks out a task. Here, one can use either periodic single frame analyses or spatio-temporal analyses that video permits. Benefits here are broad, but worker safety is the key benefit.
4. Human Activity and Advancing the Front Line – more to come on this later, but in the way an industrial engineer is tasked with understanding and optimizing various human movements within a facility, computer vision also can. This ranges from manual assembly analytics (think a time study for every single assembly cycle) to automated spaghetti diagrams to social distance tracking. Here, most recent advances in computer vision, the analysis of video streams, is a necessity. This use case is also wide ranging in terms of benefits, but the primary value levers include improved quality and higher productivity of a manufacturer’s largest asset, their workforce.

Deloitte is working with several leading Computer Vision vendors to deliver on each of the use cases listed above. However, the opportunity to help advance front line workers via computer vision and video is perhaps the most innovative and exciting. One of those leading vendors is Drishti Technologies, a Silicon Valley startup that is ushering in AI-powered production, as part of the powerful ecosystem that Deloitte has convened at The Smart Factory @ Wichita. Drishti uses AI and computer vision to deliver data and insights to the front-line production team, as well as senior leadership, to improve the quality and speed of decision-making on manual production lines. Discrete manufacturers use Drishti to improve traceability, productivity, and quality at scale. Engineers, supervisors, trainers, and line associates use Drishti’s data and associated insights to make better decisions, at the right time, faster—by augmenting people, not replacing them—and by backing every data point with video.

There are many applications where Drishti can play in the human augmentation use cases. Some of which are:

1. Productivity
   • Cycle Time Reduction: time study data on every cycle allows manufacturers to isolate the "golden runs" at the step level and democratize these best practices to reduce target time and meet takt time
   • Cycle Counts and Visibility: tracking real timeline output per station to drive visibility and transparency from the front-line to operations leadership. And, to course correct and fix issues within seconds of their happening.
2. Quality
   • Step Verification: The solution allows manufacturers to track every step an operator is taking, alerting them via an Andon light or screen if a step is missed or performed out of order.
   • Anomaly Detection: We now have insights into how long every single cycle takes, which means we can spot anomalous cycles automatically.
   • Root Cause Analysis: A video birth record of every product that has gone through an assembly station means we can look up the serial number and watch exactly how the defective product was manufactured.
3. Training
   • Standardized work adherence and “mistake-proofing:” The key to manufacturing excellence is training all line associates to follow the same “golden process” every time, and Drishti can, with its real time video analytics, ensure steps are followed in the correct order on every cycle even as they are being executed.
   • Incorporating video into daily standups for best practices and sharing improvement areas.
   • Full-fledged new operator training–the wealth of video data allows your best front-line workers to cross-train their colleagues with example videos.

To help make this real, let’s dive into an example of where a large automotive supplier used Drishti to optimize a high performing line.

One of the world’s leading lean automotive suppliers started its Drishti deployment with 12 Drishti cameras on an automotive component assembly line in Guanajuato, Mexico. The selected assembly line was already high performing, but the supplier wanted to see what Drishti could do on an optimized line.

Over a 10-week period, the auto supplier team and Drishti jointly identified a number of potential improvements. In some instances, the auto supplier’s engineers were able to verify standardized work adherence and improve the standardized work instructions to reduce process delays and microstops. “Standardized work is the foundation of our assembly operations and finding opportunities to improve adherence has the potential to significantly boost our productivity,” said the operational excellence and industrial engineering manager.

The operational excellence team used the AI-generated data from Drishti to pinpoint statistical outliers, cycles that took significantly more or less time than the others. Because Drishti can sample at up to 100%, Drishti provides a massive dataset (e.g., cycle time) that is statistically significant and unbiased. Analyzing this dataset permits the easy identification of temporal outliers, cycles that are either unusually long or short. Once identified, these cycles are sent to line supervisors to determine appropriate PDCA activities including training.    

At the conclusion of the 10-week deployment, the auto supplier saw a significant improvement in productivity on the previously assumed optimized assembly line and made the decision to expand their deployment in their factories globally.

Looking into the near future, with these capabilities enabled by significant advances in computer vision, we will see sea changes in how we work. Specifically, instead of having an industrial engineer in the plant record observations using pen and paper we can introduce a "follow the sun" workforce, where for instance, remote industrial engineers in Asia can help improve processes or root cause a problem discovered at the end of the US workday while their colleagues in the US are sleeping. The future is here today.     

To learn more about how Drishti, Deloitte and an ecosystem of innovative collaborators are showcasing the tremendous power of smart factory technologies visit The Smart Factory @ Wichita.

Authors

Don Meier
Manager
Deloitte Consulting LLP­

Steve Shepley
Principal
Deloitte Consulting LLP­

Prasad Akella
Founder and Chairman­ Drishti

Endnotes

¹ Smart Manufacturing Ecosystems: A Catalyst for Digital Transformation
² Forrester Analytics Global Business Technographics® Priorities And Journey Survey, 2019
³ Beyond the hype: delivering value with computer vision #1