The Watson IoT Platform provides a spectrum of analytics that enable you to start quickly and realize value immediately and then apply more advanced analytics as you collect more data and improve analytics skills. The analytics support a progression of capabilities to cover a wide variety of devices and assets from light bulbs and elevators to trains and aircraft. The amount of time and energy you put into understanding the data and deriving insights will vary across the different types of assets. You may monitor lights for when they burn out, but you probably won’t build predictive models to predict lighting failures, unless those lights are critical to a sensitive chemical reaction that needs to be precisely controlled!
We offer analytics to meet all your needs. You can start anywhere, but this list presents a logical progression of capabilities which allow you to get started quickly and add increasingly richer analytics to gain better insights and more value from your IoT data.
- Start quickly with real-time analytics to monitor conditions. It’s easy to get started, all you need is a connected device or sensor.
- Progress to more advanced analytics using automated machine learning algorithms which learn on the fly and generate alerts for anomalous conditions based on learned patterns in your data.
- After you’ve collected some historical data, you can use data science tools to explore your data and more advanced machine learning to build comprehensive predictive models that can forecast issues based on complex data patterns and interactions.
- Then you can leverage cognitive analytics to incorporate insights from unstructured data sources such as voice, text, video, images, and acoustics.
- And, you can extend the capabilities beyond the cloud through edge analytics which run close to the source of data.
Let’s explore these areas in more detail.
You’ve got your devices connected to the Internet of Things….now what? Visualizing your devices and monitoring for well understood conditions will allow you to automate actions to improve awareness and responsiveness to emerging situations. Many use cases require the low latency of real-time analytics and monitoring to keep equipment running smoothly, business processes flowing efficiently, and failures to a minimum. The IBM Watson IoT Platform provides these exact capabilities via an embedded rules and analytics engine called Real-Time Insights. Formerly a separate service in Bluemix, it has now been integrated into the Platform.
With a highly intuitive interface, Real-Time Insights allows end users to easily define rules that monitor raw device properties, calculated values, or real-time analytic results. When triggered, rules generate alerts and drive automated actions that can send an email or text message, create a maintenance request, or actuate a device. Real-Time Insights enables organizations to monitor equipment and operations to understand and respond to deteriorating circumstances. This leads to improved responsiveness, equipment availability, and overall efficiency.
There’s going to be data—lots of data. Machine learning allows you to automatically process that data to understand what’s important, what you should pay attention to. It helps you separate normal, expected behavior from abnormal behavior through automated analysis. It also helps you analyze large repositories of data to understand correlations and patterns among hundreds of device properties. Machine learning techniques can help businesses identify the complex interactions that foretell future events such as declining asset health, failures, and quality issues.
The simplest machine learning patterns involve anomaly detection. These automated algorithms leverage streaming data as it comes in to understand patterns and raise alerts when sensor values or device properties begin to deviate from normal operating ranges. For example, rather than hard code a high or low threshold for a temperature sensor, users can configure an anomaly detection algorithm to learn the normal operating ranges and then generate alerts when conditions vary by a meaningful amount. For more information, see this recipe showing you how to detect anomalies in streaming IoT data using the Bluemix Streaming Analytics service with the Watson IoT Platform.
However, some problems require exploration and deeper analysis through additional tools to understand patterns. For exploration and discovery, the Watson IoT Platform integrates data science tools such as IBM’s Data Science Experience and Jupyter Notebooks. With these tools, data scientists can manipulate historical IoT data, apply a vast library of prebuilt and customized analytics, and visualize results to understand the shape of the data. Insights can then be applied to real-time analytics and monitoring or used to begin building richer predictive models. For more information, see this recipe about using IBM Data Science Experience with the Watson IoT Platform.
Finally, we can leverage analytics that aggregate vast quantities of historical data to build advanced statistical models to predict outcomes. These tools are used to build predictive models that can forecast issues such as asset failures or production quality issues hours, days, or even weeks in advance. With the Watson IoT Platform, we can integrate advanced analytics such as the Bluemix Machine Learning service to build and deploy predictive models that are run against real-time data or periodically to predict future events. For more information, see this recipe about using the Bluemix Machine Learning service to predict issues with a device.
You may also have unstructured data—interested in correlating that with your machine data? Cognitive analytics are ideal for analyzing unstructured data such as video, images, audio, and text. These data sources are harder to quantify and distill down to actionable insights, and analysis of this type of data requires deep learning and training to make sense of the patterns. Cognitive analytics learn and reason from interactions with their environment rather than being explicitly programmed. They use deep learning and fuzzy logic to understand data, classify results, and make decisions based on expected outcomes. In other words, they learn and reason much like you and me!
Humans are very good at understanding these unstructured sources and applying fuzzy logic. We naturally use past experiences to classify and make judgement calls on new experiences. With Cognitive analytics, we can now train our solutions to make similar judgments and recommend the best course of action. For example, We can now enhance our solutions with new sensors and cognitive capabilities like these:
- Acoustic sensors listen to equipment, analyze sound and vibration, and tell us what is wrong with it
- Cameras watch secure areas in an airport and alert us to a security threat when someone has abandoned a piece of luggage
- A service monitors call center logs and uses text analytics to correlate words and phrases to classify similar problems in the field
- A solution reviews maintenance logs, IoT data, and repair codes for a fleet of equipment to understand the types of repairs that were performed and which were most effective
The Watson IoT Platform enables you to leverage cognitive services that can help analyze this type of data and correlate findings with machine data to provide greater context and insight about the health of equipment and operations. For more information, see my blog about Cognitive IoT and this collection of Cognitive recipes.
Analyzing data at the right time and place is crucial for making decisions about operations. You need to act on some data in near real-time, close to the source, to ensure continuous operations run smoothly…and other data needs deeper analysis to understand broader implications. On-premise solutions typically incur large up-front costs for deployment, and they don’t offer a lot of flexibility or agility in accessing and trying new analytics. As a result, much of today’s data goes un-analyzed because it is simply too much to process, perishable, too sensitive to send outside your enterprise, or the network has limited bandwidth, intermittent connectivity or is too expensive for the volume.
Edge analytics enables customers to analyze data close to the source by distributing analytics to the edge of the network to process, analyze, and respond to data where it makes the most sense. It enables customers to analyze high speed, high volume data by monitoring conditions at the edge. When something interesting happens, you can respond immediately at the edge without a round trip to a central location or the cloud. In addition, low value data such as properties that are unchanging or within normal ranges can be easily handled and filtered out right at the source while high value data can be forwarded to the cloud for deeper analysis and insights across sites and fleets.
From real-time to machine learning and cognitive to edge, the Watson IoT Platform enables a wide range of analytics. Start quickly and simply with real-time monitoring, then add automated machine learning to identify anomalous behavior. Incorporate predictive analytics using historical data and advanced statistical modeling, and add unstructured data sources and cognitive analytics for contextual insights. Finally, distribute your analytics to the edge of the network to overcome bandwidth, connectivity, security, and latency challenges. Regardless of where you are in your IoT journey, the Watson IoT Platform enables analytics to meet all your solution needs.
Here’s one final link to a summary of all of our analytics recipes. Give them a try and let us know what cool things you are doing with Watson IoT Platform Analytics!