The IoT Analytics Lifecycle – From Generating Data to Predicting the Future
After working directly with companies across a wide variety of industries that are pursuing IoT initiatives, a pattern began to emerge related to how data is handled and used to deliver value. I’ve since named this pattern: “The IoT Analytics Lifecycle” and it describes the stages through which nearly every company will pass on their IoT journey.
The lifecycle includes five stages: Generate, Collect, Analyze, React, and Predict. Almost every company implementing IoT eventually wants to reach the Predict stage. It’s usually where many of our conversations with new clients start – “Why can’t my machine just tell me it’s going to break before it does?”
In reality, however, most companies are very early in the lifecycle, typically somewhere between the Generate and Collect stages. Even companies that would be considered leaders in the digital transformation space are barely in the Analyze stage.
Identifying the technology and process required to progress further down the lifecycle is critical for the success of IoT projects. The value of your IoT data greatly increases with each stage reached.
The first step for any IoT project is to generate some data. For equipment manufacturers, this is likely telemetry data from your machines, like voltage, temperature, rpm, flow rate, or fuel level. For smart building applications, this is probably data like room occupancy, motion, office temperature, and air quality.
Many companies are much further in this stage than they may realize. Especially companies in the industrial space. Many machines already have sophisticated controllers that expose all of the sensor data that’s needed. The company only needs to tap into that existing stream of information. In other cases, the data may not yet exist and new sensors and devices need to be incorporated in order to generate the data needed to eventually create business value. For example, if your problem is to understand utilization rates of your office space, you may need to purchase and install motion sensors into each room to generate the required data.
Once the data is generated, it must be collected in a central repository that can be accessed and queried by your team. This repository is sometimes called a data warehouse or data lake.
Implementing your own data warehouse can be a challenge, especially for companies that may not have existing internal expertise in managing a large data infrastructure. Each of the major cloud providers do have a data warehouse service that can be used. Azure has an SQL Data Warehouse, Amazon Web Services has Redshift, and Google Cloud Platform has BiQuery. Although using a cloud provider does make things easier, it can still require a large amount of operational overhead to implement and maintain. IoT-specific cloud providers, like Losant, offer a fully managed data warehouse solution that eliminates all of the overhead required to store all of this generated IoT data. Which implementation direction you ultimately choose is greatly dependent on your own internal capabilities and expertise.
Now that the data is stored, it can be analyzed. This is also an inflection point on the analytics lifecycle when it comes to the value of your data. While there is some amount of value in generating and collecting data, this is the point at which you can start delivering real ROI.
Analyzing data is about understanding what happened in the past. The first step in this stage is typically visualizing your historical IoT data on a dashboard. There’s an enormous amount of value in being able to see your information for the first time. Field technicians can view this data to better understand the past behavior of a piece of equipment being repaired. Facilities managers can view historical occupancy rates to better plan office environments and services.
Analyzing data is also about processing information in different ways in order to derive insights from raw data.
The image above shows an example of how raw data can be reworked into much more valuable insights. The left graph plots the raw activity in a busy location in an office environment. Beyond seeing that the traffic drops in the evening, there’s really no useful information we can obtain. The graph on the right is populated with the same raw data, however it’s grouped by time-of-day and day-of-week. With this graph, it’s easy to see that the busiest times are Thursday and Friday at 5 p.m. This is a useful insight that a facilities manager can use to better plan for this particular office environment.
If your company can reach the React stage in the analytics lifecycle, you’ll begin to see significant value being added to your organization. The React stage is about automatically making real-time decisions from your data that feed back into a business process.
For industrial companies, reacting to data can be used to implement more efficient condition-based maintenance techniques to reduce cost. When faults are detected by your equipment, your environment can react by automatically generating a support ticket in another system. In office environments, conference rooms can be automatically freed up if no motion is detected in a booked room. In construction environments, water can be automatically shut off if flow is detected after work hours. No matter the industry, reacting to data is a fundamental part of IoT’s potential value.
This inter-system orchestration can be difficult to implement because of the wide variety of devices and protocols that have to come together. Losant adopted a visual approach to this orchestration, which we call a “workflow engine”. Compared to writing code, the visual workflow editor provides a much more intuitive environment to work in. With the ever-expanding IoT landscape, however you choose to implement the React stage, ensure it is based on a flexible and agile foundation that can keep pace with this landscape’s changes.
Predicting the future can be the end-result of successfully traversing all stages of the IoT Analytics Lifecycle. The Predict stage is about identifying the predictive indicators that lead to eventual failures.
Predictive analytics and machine learning are very misunderstood and over-hyped terms. They’re often viewed as a “cure all” to solve any problem. The reality is that it’s not magic and it requires a lot of time, data, and effort.
The ecosystem and tools around predictive analytics are still young, which means nearly every project requires a significant amount of human labor to implement. A large amount of data must be collected, often over several years, in order to properly train a predictive model. This amount of required data and time involved is why the Predict stage is last in the analytics lifecycle. As you progress through the lifecycle, each stage adds additional value while you build up the data and expertise that’s required for this final stage.
Fortunately for industrial companies, the existing equipment and the built-in fault codes they already produce are a perfect fit for supervised machine learning techniques. The goal for these companies is to use predictive analytics to predict that a fault code is about to happen based on other sensor data. Based on the data and the quality of the model, this prediction can occur with enough notice to entirely prevent downtime.
The best advice for traversing the analytics lifecycle is to start small. Identify a single, high-impact problem and begin implementing an IoT solution to solve it. Many companies fail to gain traction in their digital transformation strategy because they design an approach that’s far too large and complicated. You want to demonstrate value in months rather than years. Once you’ve experienced the value in a single project, identify the next problem and repeat the process. Most of the work you’ve put into solving it the first time, can be used again the next time.
Losant is exhibiting at IoT Tech Expo North America at booth #390. Learn more about how to engage with Losant by visiting here. You can also hear Brandon discussing this topic at the event within the Connected Industry conference track. His session ‘IoT Analytics Lifecycle: From Generating Data to Predicting the Future‘ will take place on November 28 at 11.50am.