The Internet of Things requires significant changes in the way IoT companies develop applications, and software...
developers are continuing to find it a challenge to deal with the task of enterprise-wide IoT integration. When applications will have to orchestrate data flows from devices in the field to back-end applications connected to user interfaces running on a variety of different platforms, it's a challenge.
"A lot of IoT applications have to be connected, so application integration is the second hardest problem for IoT. The first hardest problem is security," Roy Schulte, vice president at Gartner, said. A variety of IoT integration services have emerged to support this integration, but these are still in their infancy. "There is no magic backplane that natively talks to all of the applications. In fact companies will have multiple IoT platforms just like they have multiple databases today."
Getting sound insight to build IoT companies
A Gartner survey of IoT applications found that predictive maintenance was the most common application today. These applications leverage real-time sensor data about the conditions of machines in order to fix them before they fail. An early pioneer in this field is Augury, which is developing a mechanical diagnostic platform for IoT. Saar Yoskovitz, co-founder and CEO of Augury, said, "Our goal is to diagnose everything that has moving parts. The service can listen to machines and, from the sound, tell if they have a malfunction and of what type."
Augury has developed a set of vibration and ultrasonic microphones that can be affixed to any type of motor to make better sense of their health based on their sounds. The special microphone has two sensors. The vibration sensor monitors 2 Hz to 6 kHz, while the ultrasonic sensor listens from 10 kHz to 120 kHz. By comparison, traditional microphones sample from 20 Hz to 20 kHz. The vibration spectrum makes it easier to identify a mechanical imbalance and the alignment between the motor, pumps and the physical assembly of the machine. The ultrasonic sounds make it easier to listen for the sounds of metal on metal, such as a broken tooth in a gearbox.
These recordings are sent into the cloud to keep a running record of changes over time. Deep learning is used to improve algorithms that correlate machine health with different sound properties. The goal is to make it easier to diagnose and rectify potential problems before they occur.
The first generation of this new service is now being used by HVAC technicians in the field to improve with a special Bluetooth-connected microphone connected to a smartphone app. Augury plans to create permanently installed sensors that can track changes in these machines on an ongoing basis. There are about 25 million HVAC machines in the commercial buildings in the U.S. Eventually, this service could find its way into consumer fridges and washing machines to improve customer service for appliance vendors.
Improving the algorithms
One of the challenges of building these kinds of platforms lies in developing algorithms for efficiently and accurately matching sounds to the mechanical characteristics for different types of machines. "The good news is that the manufacturer and model of the machine don’t really affect how they sound," Yoskovitz said. "When you drive a car, you can hear the fan belt is squealing. It does not matter if you are driving a Mercedes or Toyota."
This makes it possible for IoT companies to develop algorithms that work for different types of machines. But Augury hopes to improve it even more by crowdsourcing data that can be compared with reports from technicians working on the different types of machines. This is correlated with data from technicians on oil-leak reports and the need to do alignment of motors, Yoskovitz said. "Over time, [we] are building the largest dictionary in the world of sounds associated with malfunctions."
Cultivating new business models
Enterprises are still hashing out how to develop effective business models for IoT applications. It is important to keep the back end flexible to make it easier to evolve the platform as new revenue models are discovered over time. More data tracking can be exciting to geeks, but boardrooms want these applications to deliver sustainable revenues.
Roy Schultevice president at Gartner
Augury is exploring several different possibilities including reducing the cost of diagnostics for commercial repair firms, improving customer outreach for appliance vendors and enabling new insurance models. The company has already lined up contracts with some of the largest HVAC repair companies in the U.S. for the on-demand diagnostic service.
Yoskovitz expects appliance makers to eventually embed low cost sensors into their washing machines and refrigerators. This would make it easier to proactively send out repair technicians or recommend upgrades when machines have reached the end of their life. "After the one-year warranty, most manufacturers lose contact with the customer," he said. "If anything goes wrong, a customer will call someone on Craigslist, and the spare parts will be Chinese knockoffs. By becoming more proactive they will build a stronger brand relationship with the customer."
Over time, this data could become provide a core value to the product vendor. They might see that some malfunctions are more likely to occur in humid environments while others are more likely to occur in the desert. This would make it possible to improve recommendations and customize equipment for different markets.
Data management considerations for IoT companies
Internet applications are rapidly shifting from content generated by humans to machine generated content. Yoskovitz said, "This will require solving new challenges ... on the backhauls of the Internet, machine learning and big data analysis. We don't want to reinvent the Internet, but to leverage methods out there today."
Augury is leveraging both Hadoop for data storage and management and the Spark analytics engine to simplify this process. AWS is used for hosting the back end. The applications are built in-house using open source libraries. Augury is doing its own integration work in-house because the nature of the data is different than typical low-bandwidth use cases. "Most of the solutions for ... IoT are designed for simple sensors with low bandwidth requirements," Yoskovitz said. "Vibration data tends to be much larger. A typical sample requires about 5 MB per 30-second recording. We needed a unique solution."
Another challenge lies in addressing privacy as forethought. For now, all machine data is stored in an anonymous manner; however, continuous monitoring services will require some connection with information about the owner of the machine. This will require enterprises to work out contracts that provide value to customers without raising privacy objections. At the same time, Augury will need to orchestrate the back end to support these agreements securely.
The continuous monitoring platform will require the development of gateway servers capable of doing some kind of analysis at the edge, so that much smaller updates would need to be sent to the cloud. They also had to develop special compression algorithms that take into account the special properties of vibration. Traditional audio compression algorithms were designed to make music sound good to consumers. Predictive maintenance oriented algorithms need to be able to shrink the data files without adversely affecting analysis.
IoT applications require enterprise architectures that scale across multiple types of infrastructure. Furthermore, these need to seamlessly blend algorithms from different types of domains. Yoskovitz said, "We found out that this market is fragmented. If you want to build an IoT solution, you have to develop things ranging from the hardware to the algorithms. If you want to build IoT companies, you need to have at least nine experts on different domains just to get things off the ground."
Moving trucks with IoT in India
Enabling IoT with embedded technology
Customizable sensors with IoT device