Google’s former HR chief, Jonathan Rosenberg recounts in his book How Google Works a story about CEO Larry Page’s response to an MBA employee’s PowerPoint product plan. Page told the MBA, “Go sit with the engineers.”
Page’s point: An effective new product plan can be created only if the engineering details are understood.
Along this line of thinking, I spoke with Joe Biron, Internet of Things CTO at PTC, because the company has staked its future on IoT, and Biron has a decade of IoT engineering experience. PTC’s mature businesses, mechanical design, and product lifecycle management are closely related to the company’s industrial IoT business, where the company is focused for growth.
In the IoT field, PTC is best known for ThingWorx, a platform for the rapid development of applications designed for smart, connected sensors, devices and products. PTC has positioned ThingWorx as the necessary picks and shovels to the miners of the IoT gold rush.
Here is a summary of ThingWorx, which excludes the complexity of connectivity and the less than clear definition of IoT to keep it short. Both subjects are worthy of separate articles.
- Connects the most of the industrial automation and industrial IoT devices. (Most is used here instead of all because the field of IoT connectivity is diverse with many specialized proprietary methods and fluid with evolving new methods.)
- Serves as a framework for business analysts to create automated processes with the connected devices using a drag-and-drop UI.
- Manages and optimizes the connected system.
- Analyzes the data acquired with analytics for IoT. Think Google Analytics, but designed for IoT devices instead of web pages and smartphones.
- Connects to cloud-based business processes.
ThingWorx is extensible with custom modules built by customers’ developers and third-party developers available through the ThingWorx Marketplace. The company partners with sensor companies such as Analog Devices and system integrators to expand the range of solutions. Further, PTC acquired Kepware for its servers and ability to connect the disparate devices and protocols that evolved during the proprietary evolution of industrial automation and that erupt regularly during this IoT gold rush.
Putting ThingWorx in the middleware box would be a mistake for a business person and, likewise, a mistake for someone with technical domain experience to imagine a half-dozen Unified Modeling Language objects (UML) and move on. Connectivity, management, analytics and cloud connectivity are the problems that innovators encounter today. Different problems will present to IoT innovators next year or two years from now. But if IoT companies such as PTC are not engaged with lead users now, a product for implementers two years from now cannot be understood and built.
Because of PTC’s IoT business, Biron understands what is hard to engineer right now, and he has a good sense of what is developing. His implementer’s perspective of the influence of the machine learning and the cloud on IoT are insightful.
Machine learning and the IoT
Machine learning is programming computers with data, lots of data instead of a programming language like Java and C#. IoT computational architectures will never be IoT devices and sensors connected to the cloud without computation resources in the control loop. The low and predictable latency constraints of control logic, especially safety-related systems, eliminate the cloud as an option. That would prevent, for example, the ability for a robot to turn off quickly because a worker’s hand is in the way. Simpler control logic can run on a microcontroller, but as the control logic becomes more complex, especially large machine learning models will need powerful multicore servers that are locally situated at the edge near the IoT control loop.
Biron used the example of detecting an anomalous condition in a robot for the purposes of preventative maintenance. In most cases, heuristics, the exact conditions that cause failure, are not understood — and not available. Machine learning, given enough data, could predict failures. Such a machine learning model would need to consume a couple thousand sensor readings per second. What Biron sees as dense multi-core edge compute node would run on a machine such as an HPE Edgeline 4000 that is ruggedized and hung on a wall.
An architecture like this has to be managed from the cloud because the administrators are usually not located at the IoT edge to manage virtual machines (VM), apply updates and monitor system operation. From Biron’s experience, the cloud will include the edge. He cited VMware’s Pulse IoT Centers, which is evolving to manage VMs and deploy systems in containers from far away, and Microsoft Azure Stack as good examples of architectures for IoT hybrid clouds that have complex edge control loops such as machine learning models.
ThingWorx is used to acquire the data to train the machine learning models. In New Jersey parlance, training these models “says easy, does hard.” Biron returned to his industrial robot example. The use cases that have convinced people that machine learning works exist in narrow domains, such as natural language, language translation, object recognition, and recommendations, because these are the domains that academic and industry research has focused on to build large corpora of training data. The application of machine learning in IoT is at an earlier stage, but product teams that are building automation systems such as industrial robots are keenly interested.
A couple years of status and failure data from thousands of robots needs to be acquired first. In most cases, it does not exist. The first step is building a schema of the training data corpus made up of status and failure data. ThingWorx can be used to capture the data, but it needs to be the right data. Data scientists and domain experts — in this case, an industrial robot expert — would collaborate. The robot may be redesigned with new or more accurate sensors to complete the schema with the right status, as well as failure events to build the corpus of training data. PTC acquired Coldlight, a machine learning team that built the Neuron automated, predictive analytics platform to solve this type of problem.
The IoT, AR connection
IoT and machine learning are inextricably linked, likewise the link between IoT and augmented reality (AR) are inextricable. The reason AR and IoT are linked is these systems will collect enormous datasets, reduced to actionable information that serve managers and workers differently. ThingWorx Studio, which was built with PTC sister company Vuforia’s AR SDK, is currently available for customer trials to create IoT context through AR.
Take, for example, an industrial plant manager and a plant maintenance technician. A plant manager walking through the plant with an AR device such as a Hololens may want his tour of reality to be overlaid with actionable information about efficiency, quality and the uptime of the line. A maintenance technician might view a machine through the camera of a smartphone overlaid with preventative maintenance information, identifying the components that need to be replaced. The technician would be guided with an AR app created by the machine maker to guide the technician through maintenance procedures.
IoT hype cycle
IoT is in the Silicon Valley hype cycle vortex. In the vortex, to those who have suspended their disbelief, IoT and its benefits can be bought today with a credit card at Fry’s or the Microcenter. IoT is a development stage technology. People like Biron take account of the problems that have been solved to deliver solutions for innovators now, as well as the hundreds of open problems that will be distilled into a dozen scientific questions that, once answered, will bring IoT to maturity. This is the path of every new technology from research through development and finally maturity.
PTC is in a very interesting position, working with 500 or so innovative customers to build working IoT systems in diverse industries. To understand IoT, stay out of the hype cycle vortex and sit with engineers like Biron.