How Machine Learning Enhances The Value Of Industrial Internet of Things
Source – forbes.com
Industrial Internet of Things (IIoT) is already revolutionizing domains such as manufacturing, automobiles and healthcare. But the real value of IIoT will be realized only when Machine Learning (ML) is applied to the sensor data. This article attempts to highlight how ML augments IIoT solutions by bringing intelligent insights.
Cloud computing has been the biggest enabler of connected devices and enterprise IoT. Cheaper storage combined with ample computing power is the key driver behind the rise of IIoT. Though it was possible to capture data from various sensors and devices, customers found it prohibitively expensive to store massive datasets. Even after sufficient storage resources were allocated, the computing horse power required to process, query and analyze these datasets was missing in the enterprise data center. Much of the available resources were allocated to data warehouses and business intelligence systems that are critical to businesses. The acceptance of cloud as an extended data center changed the equation. Industry verticals such as manufacturing, automobile, healthcare and aviation are now capturing every possible data point generated by the sensors. They are taking advantage of cloud storage, Big Data and Big Compute capabilities offered by large public cloud providers. This has been the single most important factor in accelerating IIoT adoption in enterprises.
The first generation of IIoT is all about ingesting data and analyzing it. The data points originating from sensors go through multiple stages before transforming into actionable insights. IIoT platforms include extensible data processing pipelines capable of dealing with real-time data that demands immediate attention along with data that only makes sense over a period. The pipeline responsible for processing real-time data is called as Hot Path Analytics. For example, it may be too late before the IoT platform shuts down an LPG refilling machine after detecting an unusual combination of pressure and temperature thresholds. Instead, the anomaly should be detected within milliseconds followed by an immediate action triggered by a rule. The other scenario that demands near real-time processing is healthcare. Vital statistics of the patients are monitored in real time.
As data enters the IoT platform, an ingestion layer will route a subset of that through a pipeline that is designed to deal with the real-time data points. Hot path analytics is one of the fundamental building blocks of enterprise IoT platforms.
At the heart of hot path, analytics is the rules engine that is responsible for detecting an anomaly. Enterprise IoT platforms embed a sophisticated rules engine that can dynamically evaluate complex patterns from the inbound sensor data streams. Domain experts with a thorough understanding of the schema and data format define baseline thresholds and routing logic for the rules engine. This logic acts as the critical input to the rules engine in orchestrating the flow of messages. It defines nested if-then conditions that are evaluated against every inbound data point before moving to the next stage of the data processing pipeline.
The rules engine has become the core of enterprise IoT platforms. AWS IoT includes SQL-based rules engine integrated with AWS Lambda. Amazon Kinesis Analytics, the real-time stream analytics service also comes with a rules engine. Same is the case with Azure Stream Analytics, which when combined with Azure Event Hubs delivers dynamic routing capabilities. Almost every industrial IoT platform including GE Predix, SAP Leonardo, PTC Thingworx and IBM Watson have similar rules engines.
One of the key areas of Machine Learning is finding patterns from existing dataset to group similar data points (classification) and to predict the value of future data points. Advanced algorithms related to both supervised and unsupervised ML can be used for classification and predictive analytics. Since these algorithms can learn from existing data, they can identify baseline thresholds without explicitly defining them. Since most of the IoT data is based on time-series, these algorithms can predict future values of sensors based on the historical data.
The combination of multiple ML algorithms is well-suited for replacing traditional rules engine embedded in IIoT platforms. Though domain experts are still required to define the action that needs to be taken based on the conditions, these intelligent algorithms offer increased accuracy and precision.
One of the top use cases of ML in IIoT is predictive maintenance (PdM) of devices. It is often used to detect anomalies in systems that accelerate diagnoses and root cause analysis. The algorithms will be able to predict the failure of devices by correlating and analyzing the change in the pattern. PdM can report other crucial metrics such as “remaining useful life” (RUL) of devices.
Predictive maintenance is applied in domains such as aerospace, manufacturing, automobile, transportation & logistics and supply chain. For example, in a consumer scenario, a predictive maintenance system can schedule a visit to the car service center based on a predictive model. In the aviation industry, the goal of the predictive maintenance solution is to predict the probability of an aircraft being delayed or canceled, based on relevant data sources such as maintenance history and flight route information.
Hot path analytics combined with Machine Learning will become an integral part of next generation IoT platforms. Though ML and AI cannot replace domain expertise, they certainly augment the platform to deliver better insights.