With the arrival of the factory of the future, it can be difficult to keep up with all the technological advancements that are coming to manufacturing. One innovative offering is video analytics, where a video stream is analyzed in real-time to make in-the-moment decisions. These solutions promise the opportunity to drastically improve factory efficiencies in areas that include quality assurance, process improvement and proactive plant maintenance.
Although video analytics offer many benefits, they do not come without challenges, such as support for multiple network interconnections and integration with several different applications. However, by adhering to some best practices, such as establishing a cloud presence and connecting computing systems, manufacturers can ensure that video analytics will give factory operations new insights and data that can prevent issues before they happen, and even help diagnose the root cause of the problem.
Connect all computed systems into one managed one
Most factories and manufacturing plants already have controllers connected to an industrial computer. A simple, yet essential, first step is to make these systems managed. Establishing a managed software system lays a foundation for video analytics, because all the systems will be ready for the new solution. To start, factory managers should make sure that these are monitored and up to date with all security measures and applications. Once this has been established, it will be much simpler and safer for the next steps of implementation. Factory managers should also ensure that they have the necessary building blocks in their computer system to integrate video analytics, including video ingestion applications, databases to store the images and both IT and OT network access. Once these are in place, the next step of the integration process can happen.
Establishing a cloud presence
When implementing video analytics, it is vital that the systems are able to securely deploy at least one workload to the manufacturing system that can be managed from the cloud. The cloud is where the training algorithms for the image recognition being implemented on the camera will be executed. As with any new system, there are different approaches that must be considered before integration. One such approach is to use technology that can be supported across a wide range of cloud offerings. This can include utilizing a system such as the Google Anthos hybrid cloud management solution, since this is capable of executing on a number of cloud platforms in addition to Google, such as Amazon Web Services and Microsoft Azure. Embarking on a multi-cloud strategy, enables enterprises to be more resilient to changes in the supply chain. With this multi-cloud approach, the camera capture and post-prediction actions can be deployed onto any production line and can be hosted both locally and on the cloud.
Systems must be designed to support continuous learning since the deployed equipment must remain viable for 10-plus years in order to justify the investment in technology. Image identification algorithms will continue to improve as more data is gathered and deployed systems must be capable of adapting, using the cloud to embrace the benefits of latest AI models, which are subsequently securely downloaded and implemented on the edge devices.
Another approach for using a cloud model is to extend the cloud to the mission-critical edge, where visual inspection is a much more effective solution. This involves using an IoT edge device as a communication channel between the tool on the processing floor, such as a robotic arm, and the IoT hub created at the mission critical edge.
To be more specific, the inclusion of increased intelligence in edge platforms enables decisions to be made locally on richer data sets, based on camera information fused with data captured from sensor networks. The “mission-critical” aspect refers to the need for these systems to still provide deterministic (fast) real-time responses to events irrespective of what computer workloads are running.
By establishing a well-defined communication strategy between software pieces operating at both ends of the edge connection, floor operators can ensure that video analytics will run smoothly and effectively in their operations.
Create a hybrid method of multiple hosts
Deploying video analytics can enable power and cost savings and reduce the physical footprint of electronics operating in a factory. In order to successfully take advantage of these perks, video analytics applications need to obtain data from operational technology applications, such as a security camera, running on the same platform.
In order for this process to run smoothly and safely, host applications only need to access the data that the specific application has been granted access to. Furthermore, these permissions need to be immutable. They should not be adjustable while the system is running.
When video analytics solutions are managed at the mission-critical edge, a hybrid application that incorporates both local and cloud networks must be employed. This is of particular relevance when cybersecurity is considered. The inclusion of connectivity to a cloud provides a new attack surface, as compared with a system that was implemented purely on the factory floor. The opportunity for these new systems is that the IT and OT networks are connected together. The challenge of these new systems is that the IT and OT networks are connected together. A hack could result in a serious mishap, such as access to proprietary or confidential data or, potentially worse, the modification of system functionality to convert embedded platforms into bots. Careful partitioning of systems, effectively sandboxed from each other when access to system resources are considered, limits the system access to which a hacker can gain and, therefore, limits the impact of such an attack. At no point would the system “crown jewels” be accessible to the hacker.
While cybersecurity has focused on blocking attacks, architects must plan that networks will be hacked and incorporate technologies that can detect symptoms of a breach early., coupled with a strategy to manage those systems back to a known good safe and secure state. Machine learning offers immense potential here. Foundational (protected) software can be used to recognize that change in behavior (identified as more loading on a processor core, or a different pattern of accesses to memory or IO peripherals) has occurred and flag it for appropriate action to be taken.
Although there are challenges that factory managers may face when implementing video analytics, the pros greatly outweigh the cons. Beyond the cost effectiveness of video analytics, this solution allows for better quality assurance and rapid maintenance and troubleshooting. With proper and secure implementation, video analytics can bring a factory floor into the future.
Pavan Singh is vice president of product management at Lynx Software Technologies. In his role, he is responsible for product strategy, product management and product marketing. Prior to joining Lynx, Pavan had 15+ years in product related roles in technology companies ranging from startups to established corporations. Singh is also a recognized thought leader and speaker on Internet of Things with a focus on mobility, industrial and enterprise. Pavan is a graduate of Indian Institute of Technology, Bombay and has earned a MS from Texas A&M and an MBA from the Kellogg School of Management.
