IIC Journal of Innovation 6th Edition | Page 39

Spotlight on the Industrial IoT Analytics Framework
Description Industrial analytics must satisfy a higher level of accuracy in its analytic results . Any system that interprets and acts on the results must have safeguards against undesirable and unintended physical consequence .
Industrial analytics must satisfy certain hard deadline and synchronization requirements . Near instantaneous analytic results delivered within a deterministic time window are required for reliable and high quality actions in industrial operations .
When applying industrial analytics , and interpreting and acting on the result , strong safety requirements must be in place safeguarding the wellbeing of the workers , users and the environment .
The analysis of data within an industrial system is never done without the context in which the activity and observations occur . One cannot construct meaning unless a full understanding of the process that is being executed and the states of all the equipment and its peripherals are considered to derive the true meaning of the data and create actionable information .
Industrial operations deal with the physical world and industrial analytics needs to be validated with domain-specific subject matter expertise to model the complex and causal relationships in the data . The combination of first principles , e . g . physical modeling , along with other data science statistical and machine learning capabilities , is required in many industrial use cases in order to provide accurate analytics results .
Many complex industrial systems have hierarchical tiers distributed across geographic areas . Each of these subsystems may have unique analytic requirements to support their operations . Therefore , industrial analytics must be tailored to meet the local requirements of the subsystems it supports . The requirements on timing ( avoiding long latency ) and resilience ( avoiding widespread outage of service because of faults in the network or in a centralized system ) require a distributed pattern of industrial analytics in that the analytic will be implemented close to the source of data it analyzes and to the target where its analytic outcome is needed .
Industrial analytics can be continuous or batch processes . Because of continuous execution in industrial systems , a large proportion of industrial analytics will be streaming in nature , performing analysis of live data and providing continuous flow of analytics results in support of the operations . Traditional batch-oriented analytics will still be performed either for building or improving analytic models , or for human decision-making .
For the industrial analytics to support continuous operations , the analysis of streaming data and the application of analytic outcomes must be automatic , dynamic and continuous . As the technologies in industrial analytics advance , improvements in analytic modeling e . g . through learning may also be automatic .
Analytical systems require data that has meaning and context . Unstructured data , when reported without attribution to the source and the component or system it represents , makes deriving value complex since it requires the analytics to guess or infer the meaning . Inference unnecessary adds significant uncertainty into the system . Most data can be properly attributed at the source , and if this information is communicated , it can significantly increase the success and accuracy of the analytical systems .
Table 1 : Industrial Analytics Requirements
- 38 - November 2017