By Terry Arden, CEO, LMI Technologies.
Today’s factories need to generate and process data directly from the production floor to ensure quality, drive automation, and enable customization. As a result, manufacturers are embracing Industry 4.0 concepts of networked smart devices that communicate information to drive these automated processes, as well as scheduling and just-in-time processes that produce high-quality product on demand. Smart sensors are an integral part of this factory ecosystem, where digitization and computing take place at “the edge.”
Smart sensors reduce data loads and report meaningful, high-level results that supply e-commerce systems with real-time information. This allows organizations to scale and distribute processes efficiently by leveraging advances in compute speed to reduce raw data closer to the source.
How the industry is achieving higher speeds
Chip manufacturers have already moved into the era of “acceleration,” and they recognize the solution requires a hybrid of three technologies:
- FPGA (pixel-based processing)
- GPU (array processing)
- CPU (general purpose processing)
Factories typically use 2D machine vision designed around FPGAs (built into an industrial camera) to deliver raw image data to local PCs for quality control inspection. Smart 2D cameras use a CPU and FPGA to process raw data within the device for simple inspection tasks.
As the camera resolution increases from VGA to 10-25Mp, the cycle times dramatically slow down and another level of acceleration is required. That’s where the GPU comes in.
In dedicated GPU-accelerated hardware devices, such as the NVIDIA Jetson TX2 or Intel Movidius, the decision logic part of the workload runs on an ARM CPU––which is optimized for multi-threaded performance––while the compute-intensive portion of the application runs on hundreds of GPU cores in parallel. This combination is the new paradigm of performance at “the edge,” driven by high-performance multi-core technologies.
With these three technologies (FPGA, multi-core CPU, and now massive GPU cores) the resolution and speed of machine vision systems can meet demanding factory cycle times. This additional processing capability is available to drive advances in machine vision including a move from traditional 2D vision to smart 3D vision that adds the critical dimension of measuring shape to quality control.
3D smart sensors – Advanced machine vision at “the edge”
3D smart sensors have all three of the necessary technologies required for effective acceleration, including embedded FPGA and CPU power onboard the sensor itself, and the ability to be paired with a dedicated sensor accelerator (that has GPU + CPU processing power) to provide a complete accelerated solution.
The result is massive compute power at “the edge” (i.e., where the source of data is processed to produce pass/fail results). Acceleration allows engineers to digest, process, analyze, and take action on massive amounts of manufacturing data in smaller, more manageable packets.
DNA of an accelerated smart sensor – Distributed and scalable network architecture
Smart sensor networks are built on a distributed architecture that facilitates scalability by giving process engineers various strategies to develop solutions for each manufacturing cell. Applications are implemented loading job files over the factory network. These job files configure measurements that run edge devices, which require minimum interaction with coordinating elements. As a result, accelerated data processing is achieved by preventing unnecessary or undesired uploads to servers in headquartered data centers.
Onboard data processing capability
A smart sensor not only acquires data, but processes that data and communicates control decisions to factory equipment––directly from the edge, without having to send data back to a centralized location. As a result, the sensor is able to carry out computing and storage onboard so that select applications can be executed locally at very high speeds.
When 3D smart sensors digitize and measure a target object, smaller packets of high-level data are communicated to the factory, rather than transferring raw scan data continuously for processing elsewhere. This capability alleviates pressure on network bandwidth, minimizes latency, and accelerates inspection rates.
Edge computing through acceleration improves time-to-action and reduces latencies down to milliseconds while optimizing network bandwidth. In combination with factory systems and powered by accelerated smart sensor technology, edge systems have a profound impact on industrial system performance and ultimately increase quality, flexibility, and productivity for manufacturers.