This article is about how to use a standards-based approach to deploy analytics on the edge of Industrial Internet of Things (IIoT) networks.
IIoT is here today. Wherever you look in aerospace, defense, energy production and/or transport, healthcare, manufacturing, mining, materials or smart cities the proliferation of digital devices, equipment and sensors is astounding. The challenge and the opportunity for both business and technology leadership is how to effectively and efficiently derive economic value from the data produced by these systems.
IIoT has distinct differences from more common Big Data and Analytics environments. The first is the data being produced by devices with very little compute power and memory that are located far from traditional database and computing environments. The second is the data that would be transmitted back may need to be transmitted over low-bandwidth networks that have more opportunity for degradation of data quality. The third is there are business decisions that need to be made based on the device data in real-time faster then communication can occur back to the central compute system. Fourth there may be security issues involved in transmitting some data back and forth over available networks in IIoT settings.
For the IIoT enabled enterprise the goal is to close the distance between the data, the computation and the actions that define the business processes. Or put another way organizations want to execute predictive models at the point of greatest efficiency with respect to transaction performance or cost efficiency. The way these categories of problems have been solved for business by information technology professionals for decades is by leveraging open standards.
Open standards allow for applications, data and files to be shared across multiple platforms at far lower cost then would otherwise be possible. Examples of open standards that have found their way into the common vocabulary include HTML, Java and PDF. Within any enterprise information technology environment there will be dozens if not hundreds of open standards being used on a constant basis.
Edge Intelligence is generated from the results of analytics. The billion-dollar question becomes how can open standards be leveraged to enable an IIoT Enterprise? It begins with how Data Scientists today create predictive analytics. It is common for these teams to use tools like SAS, SPSS or KNIME or languages like Python or R to create complex mathematical models which are then applied against batch and/or streaming data either “on-premise” or in the cloud.
One significant area of friction is that when these advanced analytics are created they are created in a development environment. What this means is that as written they will not be able to be deployed into a high volume production IT environment. Often when it comes time to deploy these models the Data Science teams must spend quite a bit of time working with IT to custom code these models for the specific IT environment they will be deployed in. This introduces economic opportunity costs, human resource costs, and significant delay in moving data based business forward. It is not unusual for it to take 6, 9 or 12 months for a large new model to get deployed in the enterprise.
There is a solution to this problem in that almost all models can be converted into the open industry standard Predictive Model Markup Language (PMML). PMML is a very mature open industry standard that is managed by the Data Mining Group (www.dmg.org). Zementis provides software solutions that consume, execute, optimize and scale PMML for any technology environment against batch or real-time streaming data. This includes the ability to deploy models into IIoT edge computing environments without requiring custom coding.
Zementis solutions run as a traditional application in traditional compute environments. They are also available as extremely lightweight libraries that can be deployed into edge computing environments. This solves the four challenges we outlined for creating edge intelligence:
Additional advantages to using Zementis is that Data Science teams do not have to change the tools they are using today for data mining and predictive model building. In addition, since Zementis is converting the models into an abstract logical package the model can be run at the edge or from a server whether on-premise or on the cloud or both and in many places as the demands of the business require it.