Filtering Out the Noise in IoT Data

| Blog

In a recent Information Week article, Internet of Things: The Rubber Meets the Road, Chris Taylor discusses what he sees as the three challenges associated with IoT today: automating decisions, limits of imagination, and fast data.

Taylor writes, Automating decisions has advantages and risks, in part because computing currently lacks the same discernment capabilities of the human brain. We at CaseBank agree with this statement. We believe that an important source of the discernment offered by the human brain is its ability to understand the underlying system that is producing the data, and thereby to discern signal amongst the noise.

Taylor also alludes to “limits of the imagination” creating a gap between machines (which lack the intelligence to predict the future or set goals) and the human brain (which is less adept at analyzing and applying data from hundreds of sensors).

IoT has created an abundance of data that must be sifted and analyzed. Taylor notes, “As more data is created and moved faster, finding and reacting to the right data becomes harder.” This is true. Today we have no difficulty finding patterns in data. The trouble is trying to tell if that pattern is a signal (of a failure) or simply noise. Data is increasing exponentially, but the amount of guidance is not. Equipment is not suddenly failing faster, and in fact reliability is generally improving. We are not encountering more problems, just more noise. This is why the ability to separate signal from noise in the patterns detected in the data becomes so important. Companies need to avoid false positives and duplicate alarms, because these trigger field service calls, which in turn increase overall costs.

To assess an IoT alert, the logical reasoning in CaseBank’s guided diagnostic solutions software uses prior experience with the system and any similar alerts. A troubleshooting session corresponds closely to how a human would recognize an alert as valid, and work through the problem by seeking additional information to discriminate among the possible causes.

In working with customers we have noticed codes and machine intelligence are often treated as conclusions (i.e., what is wrong), but they are actually symptoms. Relatively few machine-generated error codes are absolutely unambiguous as to their cause. One needs to look for other things to support or refute the many possible causes for each symptom.

IoT data has tremendous business value across industries, including manufacturing. But, with the influx of information comes the need to better filter out the noise, duplicates and false positives. There are certainly challenges brought forth by IoT but if managed correctly and combined with practical methods to inject an understanding of the underlying system, IoT can significantly help keep equipment up and running.

Previous Next